Kinematic signatures in reaching movements during spaceflight provide evidence that humans underestimate body mass in microgravity

eLife Assessment

The authors present a solid study in the unique conditions of weightlessness providing evidence that movements carried out in 0g are underactuated. They further provide a thorough discussion based on computational modelling to address the question as to whether the CNS underestimates mass when programming movements in weightlessness. In all cases, the persistence of the observed effects in weightlessness has important implications for theories of motor adaptation.

https://doi.org/10.7554/eLife.107472.4.sa0

Significance of the findings:

Important: Findings that have theoretical or practical implications beyond a single subfield

Landmark
Fundamental
Important
Valuable
Useful

Strength of evidence:

Solid: Methods, data and analyses broadly support the claims with only minor weaknesses

Exceptional
Compelling
Convincing
Solid
Incomplete
Inadequate

During the peer-review process the editor and reviewers write an eLife Assessment that summarises the significance of the findings reported in the article (on a scale ranging from landmark to useful) and the strength of the evidence (on a scale ranging from exceptional to inadequate). Learn more about eLife Assessments

Abstract
Introduction
Results
Discussion
Methods
Appendix 1
Appendix 2
Data availability
References
Article and author information
Metrics

Abstract

Astronauts consistently exhibit slower movements in microgravity, even during tasks requiring rapid responses. The sensorimotor mechanisms underlying this general slowing remain debated. Two hypotheses have been proposed: either the sensorimotor system adopts a conservative control strategy for safety and postural stability, or the system underestimates body mass due to reduced inputs from proprioceptive receptors. To dissociate these opinions, we studied 12 taikonauts aboard the China Space Station performing a classical hand-reaching task. Compared to their pre-flight performance and to an age-matched control group, participants showed increased movement durations and altered kinematic profiles in microgravity. Model-based analyses of motor control parameters revealed that these changes stemmed from reduced initial force generation in the feedforward control phase followed by compensatory feedback-based corrections. These findings provide support for the body mass underestimation hypothesis while being inconsistent with the strategic slowing hypothesis. Importantly, the sensory estimate of bodily property in microgravity is biased but immune from sensorimotor adaptation, calling for an extension of existing theories of motor learning.

Introduction

Successful space exploration fundamentally depends on efficient sensorimotor control under microgravity conditions, where both movement accuracy and speed are critical (Strangman et al., 2014). A consistently observed phenomenon is that astronauts exhibit slower movements during routine activities in space stations (Kubis et al., 1977). This general movement slowing persists even in controlled experimental settings where astronauts are securely fastened without concerns for postural stability or safety. Indeed, fast movements in microgravity can be disadvantageous, potentially destabilizing posture, complicating the arrest of accelerated body segments, and increasing collision risks in confined spacecraft environments (Bock, 1998; Layne et al., 2001; Paloski et al., 1992; Reschke et al., 1998; Tafforin et al., 1989). Yet even when body stability is ensured, well-practiced goal-directed actions consistently show prolonged movement durations (MDs) across various tasks, from hand pointing to visual targets (Berger et al., 1997; Bock et al., 2001; Bock et al., 2003; Mechtcheriakov et al., 2002) to step tracking with joysticks (Newman and Lathan, 1999; Sangals et al., 1999). This slowing manifests as reduced peak speeds in hand pointing (Mechtcheriakov et al., 2002; Sangals et al., 1999; Weber and Stelzer, 2022). Despite extensive documentation of this phenomenon, the underlying sensorimotor mechanisms remain unclear (Bock et al., 1996b; Weber et al., 2019), presenting a significant challenge for optimizing human performance in space.

Two primary hypotheses have emerged to explain the slowing of goal-directed actions in microgravity. The conservative control hypothesis suggests that the sensorimotor system adopts a generalized strategy prioritizing safety and postural stability over speed (Berger et al., 1997; Bock et al., 2010; Mechtcheriakov et al., 2002). This adaptation may arise from the novel challenges of movement control in microgravity, where even simple actions can propagate unexpected forces through body segments. The persistent application of this conservative strategy, as a generalized adaptation to the novel environment, could explain the movement slowing in microgravity. Alternatively, the body mass underestimation hypothesis builds on observations that humans consistently underestimate the mass of handheld objects in microgravity (Ross and Reschke, 1982), suggesting that similar perceptual biases might extend to the estimation of body segment properties (Bock et al., 1996b). Such misperception would directly impact the feedforward control of movement, potentially leading to systematic underactuation of initial movements (Elliott et al., 2010; Elliott et al., 2017).

Both hypotheses predict reduced peak speed and acceleration in microgravity, but they differ in their predictions about finer-grained movement kinematics. First, typical reaching movement exhibits a symmetrical bell-shaped speed profile, which minimizes energy expenditure while maximizing accuracy or smoothness according to optimal control principles (Flash and Hogan, 1985; Todorov, 2004). The conservative control hypothesis suggests a maintained temporal symmetry as longer durations are strategically planned for optimal performance; the peak speed and acceleration are thus delayed in time (Figure 1A). In contrast, the body mass underestimation hypothesis predicts that insufficient initial force generation, resulting from a systematic sensory bias in internal models that predict movement dynamics, would lead to earlier occurrence of peak speed and acceleration (Elliott et al., 2010; Elliott et al., 2017). Previously, the timing of these peaks has shown mixed results, with some studies reporting earlier peaks (Sangals et al., 1999), others finding delays (Fowler et al., 2008), and some showing no temporal shifts (Berger et al., 1997; Mechtcheriakov et al., 2002). Second, the hypotheses make distinct predictions about feedback-based corrections during late reaching. The conservative control hypothesis suggests that movements would remain well-planned with near-optimal execution, thus requiring minimal corrections. Instead, the mass underestimation hypothesis predicts that initial underactuation would necessitate subsequent feedback-based corrections to reach the target (Chua and Elliott, 1993; Smeets and Brenner, 1995). These corrections would manifest as additional submovements, contributing to asymmetric speed profiles (Figure 1A). Current evidence for increased feedback-based corrections in microgravity remains inconsistent: while some studies support their presence (Sangals et al., 1999), others show conflicting results (Mechtcheriakov et al., 2002) or opposite effects (Fowler et al., 2008). If present, these corrective submovements should predict MD prolongation, serving as important evidence favoring the mass underestimation account for the general slowing in microgravity.

Figure 1 with 5 supplements see all

Download asset Open asset

Model simulations revealing characteristic kinematic features of reaching movements.

(A) Speed profile predictions under two hypotheses. Under normal gravity, hand speed follows a typical profile (bold apricot lines). According to the body mass underestimation hypothesis (right panel), initial underactuation produces a lower, earlier peak speed (light red), necessitating a corrective submovement (light blue) that results in an asymmetric, prolonged profile (purple). The conservative strategy hypothesis (left panel) predicts symmetric slowing with delayed peak speed (red). (B) Two-joint arm model and its simulated effective mass. The spatial distribution of effective mass is shown by black curves (solid: normal gravity; dashed: 30% mass underestimation in microgravity). Colored lines indicate effective mass values for our three experimental target directions (45°, 90°, and 135°). (**C, F**) Representative speed and acceleration profiles simulated using effective masses from B. Profiles shown for normal (solid) and underestimated (dashed) body mass in the 90° direction demonstrate reduced, earlier peaks with mass underestimation. (**D, E, G, H**) Simulated kinematic parameters across movement directions under normal (solid) and underestimated (dashed) mass conditions. Mass underestimation consistently leads to reduced amplitude and earlier timing of speed/acceleration peaks across all directions. Model simulation details are provided in the Methods.

To help distinguish between these theories, we leveraged a unique biomechanical property of arm movements: the natural variation in effective limb mass across movement directions, known as anisotropic inertia (Mussa-Ivaldi et al., 1985). Based on a two-joint arm model (Li and Todorov, 2004), reaching movements toward targets in the frontal-medial direction involve greater effective mass compared to those in the frontal-lateral direction. Based on the motor utility theory, both the magnitude and timing of peak speed and acceleration should be systematically modulated by target direction (Shadmehr et al., 2016). Similarly, if astronauts underestimate their limb mass by approximately 30% in microgravity (Crevecoeur et al., 2014), we would expect reduced peak speed/acceleration and earlier occurrence of these peaks across all the directions. Here, we examined 12 taikonauts before, during, and after their 3- or 6-month missions on the China Space Station (CSS) and compared their performance to age-matched ground controls. Participants performed rapid reaching movements to targets specifically chosen to engage different effective masses. Our approach allows us to dissociate the effects of mass underestimation from strategic control by examining how movement kinematics vary with direction-dependent effective mass.

Results

We compared the performance of 12 taikonauts to that of age-matched ground controls across multiple test sessions (Figure 7 and Supplementary file 1). Participants performed rapid reaching movements toward one of three targets on a tablet, presented in pseudorandom order to minimize movement automation and ensure active planning for each reach. Participants were instructed to reach the target with both speed and accuracy, pausing briefly before initiating the return. The emphasized requirement of moving fast would critically test whether movement slowing is ubiquitous in microgravity. The three target directions (45°, 90°, and 135°) were chosen to vary the effective mass of the moving limb, as verified through biomechanical simulations based on the two-joint arm model (Li and Todorov, 2004; Figure 1B). Our model simulations not only demonstrated the direction-dependent variation in limb mass but also predicted how actual and misperceived limb mass would affect reaching kinematics (Figure 1C–H). These predictions were derived by combining the movement utility framework for estimating planned MD (Shadmehr et al., 2016; Figure 1—figure supplement 1A, B) with a forward optimal controller for simulating reaching trajectories (Todorov, 2005; Figure 1—figure supplement 1C, D). Notably, the model predicted that underestimating limb mass would lead to reduced and earlier peak speed/acceleration across directions (Figure 1C–H).

Movement accuracy and reaction time remain intact in microgravity

All participants successfully completed the task. Invalid trials, comprising early initiation (RT <150 ms), late initiation (RT >400 ms), or measurement failures, accounted for only 1.87% of trials in the experimental group and 0.70% in the control group. The percentage of invalid trials remained stable across the pre-, in-, and post-flight phases (Friedman’s tests, all p > 0.1).

Reaching accuracy remained high despite the stringent speed requirement of completing movements within 650 ms after target appearance (Figure 2—figure supplement 2). Analysis of endpoint error using a 3 (target direction) × 3 (phase) repeated-measures ANOVA revealed no significant differences across phases for either group. While the experimental group showed slightly better accuracy for the 45° target, this difference was minimal (0.04 cm over a 12-cm reach distance).

To introduce sufficient variability in reaction times (RTs) for assessing speed–accuracy trade-offs (Sutter et al., 2021), a beep sound was played upon the target appearance in half of the trials. The random auditory signal tended to accelerate responses, thereby widening the distribution of RTs in our dataset. The beep effectively reduced median RTs by approximately 38 ms (242 ± 49 ms vs. 281 ± 50 ms; Figure 2A; see control group data in Figure 2—figure supplement 1A). To evaluate microgravity effects, we pooled data from both beep conditions and conducted a 3 (direction) × 3 (phase) repeated-measures ANOVA. The experimental group showed significant main effects of direction (F(2,22) = 25.260, p < 0.001, partial η² = 0.697) and phase (F(2,22) = 5.820, p = 0.014, partial η² = 0.346), without interaction (F(4,44) = 2.246, p = 0.111). Post hoc analyses revealed slower RTs for the 45° direction compared to both 90° (p < 0.001, d = 1.473) and 135° (p = 0.003, d = 1.427). Contrary to the conservative-strategy hypothesis, taikonauts did not show generalized slowing in RT; they actually had faster RTs during spaceflight, incompatible with a generalized slowing strategy. Reactions were faster during the in-flight phase compared to pre-flight (p = 0.037, d = 0.803), with no significant difference between in- and post-flight phases (p = 0.127). Similar effects were observed in the control group (Figure 2—figure supplement 1A), suggesting that the reduced RTs likely reflect practice effects rather than microgravity influence. Thus, participants maintained, and even improved, their ability to rapidly initiate movements in microgravity with reaching accuracy.

Figure 2 with 6 supplements see all

Download asset Open asset

Reaction time and movement duration in the experimental group.

(A) Median reaction times plotted by experimental phase (x-axis) and target direction (different colors). Dashed lines and dot lines represent the ‘beep’ and ‘no beep’ conditions. (B) Average movement durations shown in the same format. Data are combined across beep conditions due to no significant differences between them. In both panels, error bars denote standard error across participants. Asterisks indicate significance levels for post hoc phase comparisons (*p < 0.05, **p < 0.01, ***p < 0.001).

Microgravity leads to slower movement

Although showing no decrement in movement accuracy or RT, the taikonauts exhibited prolonged MD during the in-flight phase compared to both pre- and post-flight phases (Figure 2B). A two-way repeated-measures ANOVA reveals significant main effects of direction (F(2,22) = 62.555, p < 0.001, partial η² = 0.850) and phase (F(2,22) = 6.761, p = 0.015, partial η² = 0.381), with no significant interaction (F(4,44) = 1.413, p = 0.263). Post hoc comparisons showed that MD was significantly prolonged during the in-flight phase (vs. pre-flight: p = 0.014, d = 0.866; vs. post-flight: p = 0.013, d = 0.472). MD was also significantly greater for the 90° and 135° directions—those associated with larger effective mass—compared to the 45° direction (all p < 0.001 in pairwise comparisons). In contrast, the control group showed a similar directional effect but, importantly, did not show a phase effect (Figure 2—figure supplement 1), suggesting that movement slowing is specific to spaceflight.

Microgravity leads to reduced peak acceleration/speed and advanced timing of these peaks

The initial movement exhibited signs of underactuation during spaceflight, consistent with the body mass underestimation hypothesis (Figure 2—figure supplement 6). Peak acceleration and peak speed, established measures of feedforward control (Elliott et al., 2010; Elliott et al., 2017), both showed significant decreases during the in-flight phase (Figure 3). For peak acceleration, there were significant main effects of phase (F(2,22) = 6.401, p = 0.015, partial η² = 0.368) and direction (F(2,22) = 75.516, p < 0.001, partial η² = 0.873), along with an interaction effect (F(4,44) = 4.548, p = 0.016, partial η² = 0.293). Consistent with model simulations, the 45° direction, associated with a lower effective mass, showed significantly higher peak acceleration than the other two directions (Figure 3A; all p < 0.001 in pairwise comparisons). Peak acceleration decreased significantly during spaceflight across all three directions (simple main effects: 45°, p = 0.020, η² = 0.543; 90°, p = 0.004, η² = 0.673; 135°, p = 0.005, η² = 0.657). Planned contrasts further indicated lower peak acceleration in the in-flight phase compared to pre-flight (45°, p = 0.006, η² = 0.516; 90°, p = 0.008, η² = 0.485; 135°, p = 0.003, η² = 0.560), with marginal recovery in the post-flight phase for two directions (45°, p = 0.071, η² = 0.267; 90°, p = 0.062, η² = 0.282) and significant recovery for one direction (135°, p = 0.029, η² = 0.363). In contrast, the control group displayed only the direction effect (Figure 3—figure supplement 1A, B), confirming that the underactuation was specific to microgravity.

Figure 3 with 1 supplement see all

Download asset Open asset

Magnitude changes of peak acceleration and speed during spaceflight.

(A) Peak acceleration across experimental phases and movement directions, showing systematic decreases during the in-flight phase. (B) Peak speed data presented in the same format, demonstrating parallel changes to peak acceleration. Error bars indicate standard error across participants. Statistical significance levels are shown for both overall phase comparisons (black asterisks) and planned contrasts within each direction (colored asterisks): *p < 0.05, **p < 0.01, ***p < 0.001.

Peak speed showed parallel changes during spaceflight (Figure 3B). Analysis revealed significant main effects of phase (F(2,22) = 7.043, p = 0.009, partial η² = 0.390) and direction (F(2,22) = 114.420, p < 0.001, partial η² = 0.912), without interaction (F(4,44) = 1.929, p = 0.161). Peak speed was significantly reduced during the in-flight phase compared to both pre-flight (p = 0.004, d = 0.879) and post-flight phases (p = 0.046, d = 0.529), with evidence of recovery post-flight (pre- vs. post-flight: p = 0.483, d = 0.349). Target direction systematically modulated peak speed (all pairwise comparisons, p < 0.001), with the 45° direction showing highest speeds and 135° lowest—matching predictions based on the movement utility theory. The control group showed only direction effects without phase-related changes (Figure 3—figure supplement 1B), confirming the specificity of these changes to microgravity exposure.

The critical test between the two hypotheses of movement slowing lies in whether peak acceleration and peak speed occur earlier or later in microgravity. The body mass underestimation hypothesis predicts earlier timing, whereas the conservative control strategy hypothesis predicts delayed timing. We observed an overall timing advance of these two measures. For peak acceleration time in the experimental group (Figure 4A), two-way repeated-measures ANOVAs revealed significant main effects of phase (F(2,22) = 5.189, p = 0.024, partial η² = 0.321) and direction (F(2,22) = 30.375, p < 0.001, partial η² = 0.734), along with an interaction effect (F(4,44) = 4.503, p = 0.013, partial η² = 0.290). In line with predictions, peak acceleration appeared significantly earlier in the 45° direction than other directions (45° vs. 90°, p < 0.001, d = 1.675; 45° vs. 135°, p < 0.001, d = 1.491). The phase effect emerged only for directions with higher effective mass (simple main effects, 90°: F(2,10) = 6.660, p = 0.015, η² = 0.571; 135°: F(2,10) = 8.411, p = 0.007, η² = 0.627). Planned contrasts confirmed earlier peak acceleration during spaceflight compared to post-flight (90°, p = 0.005, η² = 0.530; 135°, p = 0.004, η² = 0.537), with marginal differences versus pre-flight (135°, p = 0.065). The control group showed only directional effects without phase effect (Figure 4—figure supplement 1A).

Figure 4 with 2 supplements see all

Download asset Open asset

Temporal analysis of peak acceleration and peak speed during spaceflight.

(**A, B**) Time to peak acceleration and peak speed across experimental phases and movement directions. (**C, D**) Relative timing analysis showing these same peaks as a percentage of total movement duration, revealing more pronounced microgravity effects than absolute timing measures. Error bars indicate standard error across participants. Asterisks denote significance levels for planned phase comparisons (*p < 0.05, **p < 0.01, ***p < 0.001). Note the systematic modulation of timing by phase, particularly for high effective mass conditions.

Given that overall MD is prolonged in microgravity, a more direct way to assess the symmetry of the speed profile is to compute the relative timing of speed/acceleration peaks as a percentage of MD. Indeed, the relative timing of peak acceleration showed the same reduction in microgravity with a more pronounced effect than absolute timing (Figure 4C). A two-way repeated-measures ANOVA revealed significant main effects of phase (F(2,22) = 15.414, p < 0.001, partial η² = 0.584) and interaction (F(2,22) = 7.223, p = 0.002, partial η² = 0.396), while the main effect of direction was marginally significant (F(2,22) = 3.374, p = 0.053, partial η² = 0.235). Phase effects were significant for directions with higher effective mass (simple main effects: 90°, F(2,10) = 13.327, p = 0.002, η² = 0.727; 135°, F(2,10) = 24.574, p < 0.001, η²=0.831) but not for 45° (F(2,10) = 0.609, p = 0.563). Planned contrasts confirmed significantly earlier timing during spaceflight compared to both pre- and post-flight phases (all p < 0.002 for 90° and 135° directions). None of these effects appeared in the control group (Figure 4—figure supplement 1C).

Peak speed also showed a timing advance, but only for relative timing. For absolute timing (Figure 4B), a two-way repeated-measures ANOVA revealed a significant main effect of direction (F(2,22) = 57.428, p < 0.001, partial η² = 0.839) and a marginal interaction (F(4,44) = 2.233, p = 0.095, partial η² = 0.169), without a phase effect (F(2,22) = 0.093, p = 0.864). In contrast, relative timing analysis (Figure 4D) showed significant effects of phase (F(2,22) = 8.941, p = 0.004, partial η² = 0.448), direction (F(2,22) = 4.991, p = 0.030, partial η² = 0.312), and their interaction (F(4,44) = 4.755, p = 0.015, partial η² = 0.302). The phase effect appeared only for directions with higher effective mass (90°: F(2,10) = 17.278, p = 0.001, η² = 0.776; 135°: F(2,10) = 6.189, p = 0.018, η² = 0.553). Planned contrasts revealed significantly earlier relative timing during spaceflight versus pre-flight for both 90° (p < 0.001, η² = 0.770) and 135° (p = 0.007, η² = 0.496) directions, with significant post-flight recovery in 90° (p = 0.003, η² = 0.575). The control group showed only directional effects (Figure 4—figure supplement 1B, D). Thus, after accounting for the prolongation of MD, we observed a temporal advance of peak acceleration and speed during spaceflight.

Movement slowing is associated with corrective submovements

The feedforward component of movement, captured by our model simulations, accounts for only the initial phase of reaching. With peak acceleration and speed occurring at approximately 80 and 150 ms after movement onset, and total MD spanning 300–400 ms, participants had ample opportunity for feedback-based corrections (Dimitriou et al., 2013). These corrections, if present, would manifest as secondary submovements (Meyer et al., 1988). While optimal reaching movements typically show symmetrical speed profiles (Flash and Hogan, 1985), the presence of corrective submovements creates characteristic right-skewed speed profiles (Elliott et al., 2010; Meyer et al., 1988; Woodworth, 1899). To identify such corrections, we employed an established decomposition method (Rohrer and Hogan, 2006) to detect trials containing both primary and secondary submovements (two-peak trials; Figure 5A).

Figure 5 with 1 supplement see all

Download asset Open asset

Submovements analysis revealed changes in corrective movements during spaceflight.

(A) Decomposition of a representative two-peak speed profile, illustrating primary and corrective submovements separated by the inter-peak interval (IPI). (B) Proportion of movements showing corrective submovements across phases and directions. (C) Magnitude of feedback corrections quantified by IPI, showing direction-dependent increases in microgravity. (D) Linear relationship between feedback correction changes (ΔIPI) and movement slowing (ΔMD). Each participant (shown in different colors) contributed multiple data points from different phases and directions. (**E, F**) Primary submovement characteristics: peak speed amplitude and timing. Error bars denote standard error across participants; asterisks indicate significance (*p < 0.05, **p < 0.01, ***p < 0.001).

Analysis revealed a striking increase in corrective movements during spaceflight. The percentage of trials showing two distinct submovements increased significantly during the in-flight phase (Figure 5B), evidenced by a main effect of phase (F(2,22) = 13.969, p < 0.001, partial η² = 0.559) and a marginal phase-by-direction interaction (F(2,22) = 2.642, p = 0.054, partial η² = 0.194). This effect was specific to microgravity exposure, as the control group showed no significant changes across phases (all p > 0.05; Figure 5—figure supplement 1A). Notably, the increase in corrective movements was most pronounced for directions with higher effective mass (simple main effects: 90°, F(2,10) = 11.426, p = 0.003, η² = 0.696; 135°, F(2,10) = 7.292, p = 0.011, η² = 0.593).

To understand how feedback corrections contribute to movement slowing, we analyzed the inter-peak interval (IPI) between primary and corrective submovements, an established measure of feedback-based control (Craik, 1947; Meyer et al., 2018; Meyer et al., 1988). The taikonauts showed markedly longer IPIs during spaceflight (Figure 5C), with a two-way ANOVA revealing significant effects of phase (F(2,22) = 11.150, p = 0.001, partial η² = 0.503), direction (F(2,22) = 4.721, p = 0.026, partial η² = 0.300), and their interaction (F(2,22) = 6.559, p = 0.002, partial η² = 0.374). Critically, IPI increases were confined to directions with greater effective mass (90°, F(2,10) = 11.573, p = 0.002, η² = 0.698; 135°, F(2,10) = 13.371, p = 0.001, η² = 0.728). The control group exhibited only directional effects (F(2,22) = 19.277, p < 0.001, partial η² = 0.637; Figure 5—figure supplement 1B) without phase-related changes (all p > 0.13).

To establish whether these enlarged corrective movements explain the overall movement slowing, we examined how changes in IPI (ΔIPI) predict changes in movement duration (ΔMD) during spaceflight (Figure 5D). A linear mixed model (LMM) analysis incorporating ΔIPI, phase transition (pre-to-in and post-to-in), and direction revealed that ΔIPI significantly predicted ΔMD (β = 0.458 [0.144, 0.771], t(67) = 2.915, p = 0.005, partial-R² = 0.103). Phase transition also showed significance (β = −0.01 [–0.017, –0.001], t(67) = –2.696, p = 0.009, partial-R² = 0.086), with stronger effects for pre-to-in versus post-to-in transitions, indicating incomplete recovery post-flight. Movement direction showed no significant effect (all p > 0.42), suggesting that the relationship between corrective movements and duration slowing generalizes across reaching directions.

The analysis of primary submovements revealed even stronger evidence for feedforward control changes in microgravity. By focusing on just the primary submovement in two-peak trials, we found more pronounced effects than when analyzing the overall movement (comparing Figure 5E to Figure 5—figure supplement 1C). The primary submovement showed a robust reduction in peak speed during spaceflight (main effect of phase: F(2,22) = 10.363, p = 0.001, partial η² = 0.485) that was consistent across directions (interaction: F(4,44) = 0.382, p = 0.711; Figure 5E). More tellingly, while the overall movement showed timing changes only in relative measures, the primary submovement exhibited earlier peak speed timing in absolute terms (phase-by-direction interaction: F(2,22) = 4.264, p = 0.012, partial η² = 0.279). This temporal advance was most evident in directions with higher effective mass (planned contrasts vs. post-flight: 135°: p = 0.008, η² = 0.615; 90°: p = 0.076, η² = 0.403) but absent in the 45° direction (p = 0.188, η² = 0.284). The selective nature of the primary submovement in both magnitude and timing (comparing Figures 4B and 5F) provides compelling evidence for mass-dependent changes in feedforward control. The control group showed only the expected directional variations (F(2,22) = 46.259, p < 0.001, partial η²=0.808; Figure 5—figure supplement 1D) without any phase effects (all p > 0.13), confirming these changes as specific to microgravity exposure.

The microgravity effects on movement kinematics are directionally dependent

Reaching movements are inherently directionally dependent due to anisotropies in limb dynamics (Gordon et al., 1994; Shadmehr et al., 2016). Even before microgravity exposure, key kinematic variables—peak speed, peak acceleration, and their timing—exhibited consistent directional differences, as shown in Figure 1C–H. This baseline direction-dependency suggests that effective mass varies across movement directions and influences movement execution under normal gravity. Building on this observation, we tested whether microgravity-induced changes would also follow a direction-specific pattern, as predicted by the mass underestimation hypothesis.

To assess microgravity-induced changes, we quantified the directional differences (Δ) in peak speed, peak acceleration, and their timing, comparing in-flight measurements to pre-flight baselines. The comparison between in- and post-flight was not included in this analysis since post-flight recovery was not complete (see peak speed and peak acceleration changes in Figure 3). As shown in Figure 6, these directional effects were evident across all kinematic metrics. For both Δ peak speed and Δ peak acceleration, the magnitude of change followed a consistent ranking across directions, with the largest effects at 45°, intermediate at 90°, and smallest at 135° (i.e., 45° > 90° > 135°). Similarly, for timing metrics—Δ peak speed time and Δ peak acceleration time—the shortest delays occurred at 45°, while 90° and 135° showed comparably larger shifts. Note the 45° did not show significant timing advance, as previously shown in Figure 5F.

Figure 6

Download asset Open asset

Direction-dependent effects of microgravity on peak kinematics and their timing.

Left: model outputs. Right: experimental data shown as Δ relative to the in-flight session; solid bars = Δ(in − pre) and semi-transparent bars = Δ(in − post). Colors encode direction consistently across panels (e.g., 45° = darker hue, 90° = medium, 135° = lighter/orange). Panels (clockwise from top-left): Δ peak speed (cm/s), Δ peak speed time (ms), Δ peak acceleration time (ms), and Δ peak acceleration (cm/s²). Bars are group means; error bars denote standard error across participants.

Overall, the direction-dependent changes were roughly consistent with model predictions based on the mass underestimation hypothesis (Figure 6, comparing left and right panels). The model simulation similarly indicates that amplitude changes (Δ peak speed, Δ peak acceleration) and timing advances are rank ordered by movement directions. While the absolute magnitudes differed between model and data, the directional trends were broadly aligned. To assess the overall consistency between model predictions and experimental measurements, we performed repeated-measures correlation analyses between simulated and observed Δ values across directions (Bakdash and Marusich, 2017). Significant within-subject correlations were found for Δ peak speed time (rₘ = 0.627, p < 0.001), Δ peak acceleration time (rₘ = 0.591, p = 0.002), and Δ peak acceleration (rₘ = 0.573, p = 0.003), while Δ peak speed did not reach significance (rₘ = 0.334, p = 0.103). Together, these results suggest that the microgravity effect is direction-specific and the rank ordering of directions is broadly aligned with the predictions of the mass underestimation model. We note that these correlations evaluate the directional trend rather than the absolute magnitude of the effects; a precise quantitative match is not expected given the simplifications of the two-joint arm model.

Speed–accuracy trade-offs remain intact during spaceflight

The speed–accuracy trade-off, a fundamental indicator of motor control efficiency, remained largely intact during spaceflight. We evaluated this trade-off using two complementary approaches. First, following the traditional Fitts’ framework, we examined the relationship between MD and endpoint dispersion, which was not negatively affected by microgravity (Figure 2—figure supplement 3). Second, we analyzed the trade-off between RT and initial movement variance (Figure 2—figure supplement 4), which reflects the quality of movement planning (Sutter et al., 2021). This relationship persisted across all phases in both groups. Thus, while microgravity altered movement execution through mass underestimation, it did not fundamentally disrupt the sensorimotor system’s capacity to regulate the speed–accuracy trade-off. Additionally, in both groups, we found a significant negative correlation between MD and RT, both across and within individuals (Figure 2—figure supplement 5). This finding indicates that participants moved faster when their RT was slower, and vice versa. This flexible motor adjustment, likely in response to the task requirement for rapid movements, remained consistent during spaceflight.

Discussion

Long-term exposure to microgravity presents a unique environment that cannot be replicated on Earth, offering invaluable insights into human sensorimotor adaptation. Understanding how microgravity affects motor performance is crucial both for ensuring successful space exploration and for advancing our fundamental knowledge of motor control principles. Our study investigates the mechanisms underlying one of the most distinctive motor signatures in microgravity: the slowing of goal-directed actions during spaceflight. While taikonauts maintained reaching accuracy and rapid RTs, their movements exhibited systematic changes in microgravity characterized by reduced peak speed and acceleration, together with direction-dependent advances in peak timing, particularly for movements involving higher effective mass. These persistent kinematic alterations (Figure 4—figure supplement 2) are not readily explained by a simple conservative-control account alone. Instead, they are consistent with an underactuation pattern that could arise, at least in part, from an underestimation of limb effective mass in microgravity.

These changes were broadly consistent with model predictions across movement directions with varying effective masses due to limb biomechanics, suggesting that the representation of body mass may affect movement control during spaceflight. Beyond the initial underactuation associated with altered feedforward control, we observed increased feedback-based corrective submovements that predicted MD prolongation in spaceflight. These effects were specific to microgravity exposure, occurring only during spaceflight with marked recovery post-flight, while ground controls showed no comparable changes. Notably, the taikonauts’ fundamental motor control capabilities remained largely intact, as evidenced by preserved movement accuracy, RT, and speed–accuracy trade-offs.

The underactuation observed in initial movements under microgravity extends findings from parabolic flight studies (Crevecoeur et al., 2014; Papaxanthis et al., 2005). When individuals performed reaching movements with handheld weights during parabolic flight-induced gravity alterations, their reaching trajectories and grip forces showed systematic changes suggesting mass misperception (Bock, 1998; Bock et al., 1996a; Bock et al., 1996b; Crevecoeur et al., 2014; Papaxanthis et al., 2005). In zero-gravity conditions, goal-directed arm movements were slower (Papaxanthis et al., 2005), and grip-load force coordination was altered (Crevecoeur et al., 2010; Crevecoeur et al., 2014), consistent with mass underestimation. Conversely, hypergravity (1.8 g) induced increased peak acceleration and speed in reaching movements, along with elevated grip force during object manipulation, suggesting mass overestimation (Bock, 1998; Bock et al., 1996b; Bock et al., 1996a; Bock et al., 1992). Notably, modified grip-load force coordination reflects changes in feedforward control (Flanagan and Wing, 1997; Johansson and Edin, 1993), indicating that mass misestimation influences action planning. However, while parabolic flight studies capture only transient adaptations to rapid gravity transitions, our study targets general slowing during spaceflight and reveals the enduring effects of sustained microgravity exposure.

We also considered alternative explanations beyond mass underestimation. First, we tested whether a change in the cost-function of our optimal control model, which simulates a more conservative control, could account for movement slowing. This was implemented as a uniform rescaling of the state and control weights (Q and R) in the cost function (Crevecoeur et al., 2010). Increasing the scaling factors (α) will simulate the increase in control effort penalty and a more conservative control. However, while this manipulation reduced peak velocity and acceleration, it also delayed their occurrence, a pattern opposite to the observed effect. Thus, a uniform change in the cost function alone does not reproduce the observed combination of reduced peak amplitudes and advanced peak timing. Details of the cost-function simulations are provided in Appendix 1 and Figure 1—figure supplements 2 and 3. Second, changes in neuromuscular properties may also contribute to the observed underactuation. In microgravity, tonic muscle activity is likely reduced due to the absence of sustained antigravity demands, and descending vestibular inputs to motor neurons are diminished (Fisk et al., 1993). These changes could lower motor neuron excitability and alter muscle activation dynamics, potentially resulting in a weaker and slower initial response to an otherwise equivalent motor command—even in the absence of mass misestimation. While our simplified model treats the actuators as ideal torque generators and therefore does not capture these neuromuscular factors, such factors alone do not straightforwardly explain the direction-dependent pattern observed here. In the absence of additional direction-specific mechanisms, changes in tonic muscle activity, vestibular drive, or reflex gain would be expected to influence movement more generally across directions. The stronger effects observed in directions involving higher effective mass are therefore more naturally captured by a misrepresentation of inertial properties, although neuromuscular changes may also have contributed to the overall underactuation. Third, muscle weakness, a common effect during prolonged spaceflight, might also contribute to movement slowing. However, recent large-sample studies suggest that muscle strength reduction is mainly in lower-limb muscles that counter gravitational pull, not in upper-body muscles (Scott et al., 2023). Furthermore, the torques required in our task (~2 N m for ~12 cm reaches) are relatively small, making it unlikely that muscle weakness is a primary factor of the observed kinematics, though a minor contribution cannot be excluded. Fourth, finger-screen friction might slow down the reaching movements, particularly in microgravity where astronauts must actively press on the screen to maintain contact in the absence of gravitational loading on the hand. However, during typical interaction with a touch screen, frictional forces are modest, generally ranging from 0.1 to 0.5 N (Ayyildiz et al., 2018), and directional variations are likely even smaller. These values are minimal compared to the 10–15 N required to accelerate the arm during reaching, making frictional anisotropy an unlikely source of the directional kinematic changes. Fifth, body stabilization in microgravity might also contribute to movement slowing in space. In fact, our participants were securely strapped at the feet and used the left hand to grasp a fixed bar for support, providing multi-point stabilization of the body. If trunk displacement were the primary source of the observed effects, it would be expected to uniformly attenuate all kinematic measures, particularly in directions where the reaction force exerts a larger torque on the trunk. However, the 45° direction—where trunk perturbation is expected to be minimal according to this account—showed significant changes in MD, peak acceleration, and peak speed, while not showing significant timing changes. This dissociation suggests that trunk instability alone cannot account for the observed pattern. Nonetheless, we did not directly measure trunk or shoulder kinematics during the experiment, and we cannot entirely rule out that small trunk displacements contributed to some extent. Future studies would benefit from recording trunk and shoulder motion to more definitively address this factor. Finally, Coriolis and centripetal torques, omitted from our simplified model, might also contribute to movement slowing or direction effect we observed. Previous work has suggested that such effects may be non-negligible in fast planar reaching (Hollerbach and Flash, 1982). To examine their potential contribution, we incorporated Coriolis and centripetal torques terms in our extended model (see Appendix 2 and Figure 1—figure supplements 4 and 5 for details). These torques were also directionally anisotropic, but their magnitudes were too small to meaningfully affect movement utility or effort computation in our model. Thus, Coriolis and centripetal torques are unlikely to explain the large and systematic effects we observed.

Our results challenge a simple uniform conservative-control account, which posits that movement slowing reflects a generalized adaptation to address stability or safety concerns during spaceflight (Bock, 1998). Arguing against a purely strategic account, taikonauts demonstrated faster RTs during spaceflight—inconsistent with a generalized slowing strategy. Furthermore, if the sensorimotor system adopted a conservative strategy with pre-planned longer MDs, the speed profile should maintain symmetry with delayed peak acceleration and speed. Instead, we observed left-skewed speed profiles with earlier peak occurrence for the 90° and 135° directions, where effective mass is greater. This pattern is more consistent with mass-dependent underactuation than with a uniform strategic adjustment, although we note that the 45° direction did not show a significant timing advance, leaving open the possibility that strategic factors may also play a role in some conditions. This temporal change was more pronounced when the primary submovement was isolated to pinpoint the feedforward control of the reaching since the overall speed profile is affected by the later feedback-based corrections. Previous studies reported inconsistent findings regarding speed profile symmetry, including advanced (Sangals et al., 1999), delayed (Fowler et al., 2008), or unchanged (Berger et al., 1997; Mechtcheriakov et al., 2002) peak speed timing during spaceflight. These discrepancies likely reflect methodological differences, including task variations and technical limitations. For example, two studies used joystick-based pointing tasks without involving limb movement (Fowler et al., 2008; Sangals et al., 1999), which arguably makes them less suitable for examining mass underestimation effects. Mechtcheriakov and colleagues used an outstretched arm task, but their limited sampling rate of 25 Hz may have hindered precise quantification of peak speed, preventing submovement analysis. Additionally, our larger sample size (n = 12) may have enhanced detection of microgravity effects compared to previous studies with typically three to seven participants (Berger et al., 1997; Bock et al., 2001; Bock et al., 2003; Bock et al., 2010; Fowler et al., 2008; Mechtcheriakov et al., 2002; Papaxanthis et al., 1998; Weber and Stelzer, 2022) or single-participant designs (Heuer et al., 2003; Manzey et al., 1998; Sangals et al., 1999).

If mass underestimation contributes to the observed underactuation, it likely arises through two complementary mechanisms: degraded proprioceptive feedback and misinterpretation of weight-related cues. Microgravity impairs proprioception by reducing both muscle spindle sensitivity (Lackner and DiZio, 1992) and joint receptor responsiveness (Proske and Weber, 2023)—inputs that are crucial for weight perception (Proske and Gandevia, 2012). Furthermore, the sensorimotor system may interpret weight unloading as reduced body mass rather than environmental change (Bock et al., 1996b). Recent studies demonstrate that the sensorimotor system probabilistically infers body-versus-environment causality based on available sensory cues (Berniker and Kording, 2008; Fercho and Baugh, 2014; Kluzik et al., 2008; Kong et al., 2017; Wilke et al., 2013). In normal gravity, consistent mass–weight relationships lead to heavy reliance on weight-related cues for mass estimation (Bock, 1998; Bock et al., 1996b; Sangals et al., 1999). In microgravity, however, reduced weight cues become misleading and may systematically bias mass perception, affecting feedforward control of movement. We note, however, that the same proprioceptive degradation could also affect motor output through other pathways—for instance, by reducing tonic muscle activation or altering spinal reflex gains (Fisk et al., 1993)—independent of any explicit misrepresentation of body mass. Disentangling these mechanisms will require future experiments that can independently manipulate proprioceptive loading and limb inertia.

The persistence of deviant feedforward control and the subsequent within-movement corrections appears intriguing from the theories of human motor learning. Humans demonstrate remarkable ability to adapt to environmental changes in terrestrial studies. For experimentally applied visual or force perturbations, individuals can recalibrate their sensorimotor control and restore baseline performance within minutes (Krakauer et al., 2019), a process often theorized as forming internal models of the new environment (Kawato, 1999; Wolpert et al., 1995). Relatedly, even when the body mass is perturbed by attaching a weight on the forearm, individuals can fully adapt their reaching movements within tens of trials (Wang and Sainburg, 2004). Previous studies on parabolic flights have also shown that the control of arm movements exhibited rapid adaptation to gravitational changes over the course of several parabolas (e.g., Gaveau et al., 2016). In contrast, our in-flight sessions were scheduled at least 3 weeks post-launch—even well beyond the 2–3 weeks window typically considered critical for sensorimotor adaptation in spaceflight (Kanas and Manzey, 2008), but the participants still exhibited unadapted feedforward control and internal model, suggesting that the underlying sensory bias was not corrected through experience.

We believe this contrast stems from distinct natures of perturbation effects in microgravity when compared to in terrestrial environments. Previous studies on sensorimotor adaptation simulated environmental changes by imposing a perturbation to alter the relationship between motor commands and sensory consequences of motor commands. For reaching adaptation here, visuomotor perturbations change the spatial mapping between hand motion and its visual representation’s motion (Krakauer et al., 2000; Wolpert et al., 2011); force perturbations change the dynamic mapping between actual force outputs and hand motion (Lackner and Dizio, 1994; Shadmehr and Mussa-Ivaldi, 1994). These mapping changes, though novel, are consistent; thus, the sensorimotor system is capable of approximating them and adapting accordingly. Theories of sensorimotor adaptation conceptualize the formation of the new mapping as acquiring internal models, which link motor command and efference copy to sensory consequences of motor commands (Flanagan and Wing, 1997; Kurtzer et al., 2008; Wolpert et al., 1998; Wolpert and Miall, 1996). However, the novel environment of microgravity may alter the sensory estimate of motor apparatus (i.e., the body mass here), not merely the sensorimotor mapping. In normal gravity, weight-related sensory cues—mostly proprioceptive feedback—accurately represent body mass. In microgravity, however, these weight-related cues are substantially reduced due to the absence of gravitational pull, providing persistently biased information about body mass. In other words, the microgravity environment failed to provide veridical sensory cues for mass estimation, and reduced gravitational and proprioceptive cues persistently misinform the controller about body mass. Since body mass is a parametric input to the internal models, its bias would affect motor actions even when the internal models are well adapted. Hence, the unique environment of microgravity may reveal an important constraint of the sensorimotor system, that is, its quick adaptability is restricted to learning a novel sensorimotor mapping, not to persistent sensory bias of bodily property.

The discrete nature of our reaching task may contribute to the persistence of sensory bias. Within-movement corrections indicate that the sensorimotor system learns about sensory bias of mass estimation (Dimitriou et al., 2013), possibly via kinesthetic feedback during movement (Papaxanthis et al., 2005; Proske and Weber, 2023; Weber and Stelzer, 2022). However, movement-related cues only become available after movement initiation. Once a movement ends and the hand returns to rest, microgravity continues to elicit misleading sensory cues that bias mass estimation between trials. This lack of between-trial adaptation parallels findings from a manual interception study aboard the International Space Station, where astronauts persistently initiated movements too early when attempting to intercept an object moving at constant speed (McIntyre et al., 2001). Although astronauts corrected their movements online within each trial, their initial timing error persisted across multiple trials and sessions during a 15-day spaceflight. Hence, discrete motor tasks, such as manual interception and reaching, rely heavily on feedforward control and are susceptible to sensory biases induced by microgravity. This would also be consistent with the observation that continuous reaching between targets, without stopping, shows quicker adaptation to zero gravity in parabolic flights (Papaxanthis et al., 2005) and does not exhibit prolonged MD during spaceflight (Fowler et al., 2008).

What are the potential consequences of the underactuation observed in microgravity beyond the simple reaching movements examined here? The implications may be particularly significant given two key considerations. First, the majority of human movements are not deliberately controlled (Brooks, 1986; Marsden et al., 1976), suggesting that underactuated initial movements may be pervasive in microgravity—especially for actions involving larger body masses such as whole-body movements or object manipulation. Without the controlled conditions and explicit speed requirements of our experimental task, such underactuation might manifest even more prominently in natural movements during spaceflight. Second, our finding of increased reliance on feedback-based corrections points to a potentially costly compensatory mechanism. As feedforward control becomes systematically underactuated, whether due to mass underestimation, altered neuromuscular properties, or both, the sensorimotor system must depend more heavily on feedback control to achieve adequate performance. This increased reliance on feedback processes may demand additional cognitive resources for action regulation (Bock et al., 2010), potentially contributing to the nonspecific stressors that have been shown to impair human performance in space (Tian et al., 2024). The cognitive cost of these compensatory adjustments might be particularly relevant for complex motor tasks or situations requiring divided attention during spaceflight operations.

The preservation of speed–accuracy trade-offs in our study extends previous findings from Fitts' task experiments performed during spaceflight (Fowler et al., 2008). Both the traditional speed–accuracy trade-off and action planning-dependent trade-off remained intact in microgravity (Figure 2—figure supplements 3 and 4), suggesting preserved fundamental motor control capacity. Motor performance during spaceflight typically degrades in tasks with high cognitive demands (Eddy et al., 1998; Manzey et al., 1998; Strangman et al., 2014), particularly those requiring sustained attention (Tian et al., 2024). Our findings suggest that while microgravity-induced underactuation affects movement execution, the underlying motor control capabilities remain robust in microgravity when cognitive demands are modest.

Our study presents four methodological limitations that warrant consideration. First, despite matching for age and gender, our ground controls exhibited faster movements than the taikonauts. This systematic difference, reflected in higher peak acceleration and speed, may stem from differing experience with the touchpad device used for movement recording. While taikonauts used touchpads regularly, 9 of 12 control participants had minimal experience, potentially leading to compensatory faster movements to avoid overtime errors. Second, the temporal evolution of mass underestimation effects remains unclear. Our earliest measurements occurred 3 weeks post-launch, though parabolic flight studies suggest these effects may emerge within hours of microgravity exposure (Crevecoeur et al., 2014; Papaxanthis et al., 2005). Additionally, incomplete post-flight recovery in some kinematic measures suggests either a mass overestimation due to exaggerated weight-related cues or neuromuscular deficiency typically observed shortly after returning to Earth (Tays et al., 2021). Third, while our hand-reaching task enabled us to use well-established analytical frameworks for studying motor control aspects of movements, the generalizability of mass underestimation effects to other movements, particularly whole-body actions, remains to be determined. Future research should address these limitations by implementing matched device training protocols, adding early- and post-flight measurements, and extending investigations to diverse motor tasks. We also notice that the microgravity effect was less consistent for the 45° reaching direction. This direction engages more single-joint, elbow-dominant reaches, whereas the 90° and 135° targets require greater multi-joint (shoulder + elbow) coordination. Consistent with this, an exploratory analysis of hand-path curvature (cumulative curvature; positive = counterclockwise) showed larger curvature at 45° (6.484° ± 0.841°) than at 90° (1.539° ± 0.462°) or 135° (2.819°±0.538°). The significantly larger curvature in the 45° condition suggests that these movements deviate more from a straight-line path, supporting a more elbow-dominant movement in this direction. Importantly, this curvature pattern was present in both the pre- and in-flight phases, indicating that it is a general movement characteristic rather than a microgravity-induced effect. We postulate that single-joint movements, more so than we considered in our ideal two-link model, made effective mass rather small in the 45° direction; hence, the mass underestimation and underactuation in feedforward planning had smaller effect, compared to model simulations, in this direction. Fourth, our interpretation of the observed underactuation relies on a simplified optimal control model that treats muscles as ideal torque generators. This model does not capture several neurophysiological factors that may be altered in microgravity, including changes in tonic muscle activation, spinal reflex modulation, and descending vestibular influences on motor neuron excitability (Fisk et al., 1993), nor potential alterations in the damping and natural frequency of the limb’s mechanical response. However, the available evidence suggests that upper limb muscle capacity is relatively preserved in microgravity: a systematic review found that upper limb maximal voluntary contraction remained mostly unchanged during unloading periods of up to 45 days, with upper limb muscles declining substantially more slowly than lower limb and trunk muscles (Winnard et al., 2019; Bosutti et al., 2025). While the direction-dependent pattern of our results is most parsimoniously explained by a misrepresentation of inertial properties, we cannot exclude that some portion of the observed underactuation arises from these other neuromuscular factors. Distinguishing between mass underestimation and other sources of underactuation would benefit from future studies that combine detailed musculoskeletal modeling with direct measurements of muscle activation, joint impedance, and trunk kinematics during spaceflight.

In conclusion, our study provides converging evidence that movement slowing in microgravity is associated with systematic underactuation of initial movements, a pattern most consistent with the underestimation of body mass hypothesis. By leveraging the direction-dependent variation in limb effective mass, we showed that reaching movements in microgravity exhibit kinematic signatures—reduced and earlier peak speed/acceleration—that are broadly consistent with the predictions of the mass underestimation hypothesis. Although other neurophysiological factors such as altered muscle tone and vestibular modulation may also contribute, these effects are not readily explained by a uniform strategic adjustment or general neuromuscular change alone. The persistence of these effects throughout long-duration spaceflight, coupled with increased feedback-based corrections, suggests how altered sensory inputs can impact both predictive and reactive components of motor control in ways that resist sensorimotor adaptation. While basic motor control capabilities remain intact, as evidenced by preserved speed–accuracy trade-offs, the consistent underactuation of movements may have significant implications for the theorization of sensorimotor learning and for practical considerations of motor performance during space operations. These findings not only advance our understanding of how the sensorimotor system adapts to novel gravitational environments but also highlight fundamental principles of sensorimotor integration, particularly the central role of gravitational cues in shaping our perception of body dynamics.

Methods

Participants

Twelve taikonauts (2 females, 10 males; mean age 49.5 ± 6.5 years) from the first four missions of the CSS served as the experimental group (Supplementary file 1). The control group consisted of 12 right-handed residents from Beijing (2 females, 10 males; mean age 49.9 ± 6.5 years). The control group received monetary compensation. All participants were provided with written informed consent forms, approved by the Ethics Committee of the China Astronaut Research and Training Center and the Institutional Review Board of Peking University. The experimental group was exposed to microgravity for 92–187 days during manned missions in China’s Shenzhou Program. Due to operational constraints, the number of test sessions varied among taikonauts: one completed 4 sessions, five completed 6, and six completed 7 (Supplementary file 1). All control sessions were conducted at Peking University.

Task and experimental design

The task was conducted using a tablet (Surface Pro 6, Microsoft, Redmond, Washington) with 100 Hz sampling frequency and noise-canceling earphones (FiiO FA7, FiiO Electronics Technology, Guangzhou, China). During in-flight sessions, participants wore earphones and assumed a neutral posture with feet secured by foot straps in front of a foldable tabletop, positioned 0.9 m above the cabin floor. They held the tabletop edge with their left hand, allowing their right hand to move freely on the tablet (Figure 7A). The foot straps and left-hand grip provided body stabilization. The tablet, attached to the tabletop with Velcro, was placed ~35 cm in front of participants and 30 cm below chest level (Figure 7A). At the start of each trial, an orange circle (0.5 cm radius) appeared at the bottom of the screen, signaling participants to move their right index finger to this start circle. After a random delay (500–1000 ms), a target appeared at one of three possible locations, 12 cm from the start circle at 45°, 90°, or 135° counterclockwise from the horizontal axis (Figure 7A). Participants were instructed to reach the target quickly and accurately, pausing briefly before returning to the start. If the time from target appearance to movement end (finger speed <10 cm/s) exceeded 650 ms, a ‘too slow’ message prompted faster movement. In half of the trials, a 50-ms ‘beep’ sound was played at target appearance. Each session included 120 trials, with 3 target directions and 2 beep conditions balanced and randomized (Figure 7B). The x–y position of the index finger was recorded continuously at 100 Hz.

Figure 7

Download asset Open asset

Experimental setup and design.

(A) Top-down view of a participant performing the reaching task with the right hand on a tablet. The start position and all possible target locations are shown as orange dots on the tablet screen. (B) Experimental design. Both groups completed 4–7 sessions, with each session consisting of 120 trials. Some taikonauts missed one or two in-flight sessions.

All experiments aboard the CSS were monitored in real time by experimenters at the Beijing Control Center. Task instructions were displayed on the opening screen of the data-acquisition application, and participants were given ample time to read them. The same team of experimenters administered all pre-, in-, and post-flight sessions using identical instructions. Consistent with standard practice, the astronauts served as both participants and on-orbit experimenters and were extensively trained for this role on the ground. Multiple pre-flight sessions were conducted to familiarize them with the task. These safeguards were implemented to ensure high-quality data.

The control group was included to assess a potential confounding effect of repeated measurements. The interval between successive test sessions was 6–9 days, which was shorter than those of the taikonauts (1–2 months) to conservatively evaluate potential practice effects. Despite the shorter intervals, no significant differences were observed across sessions for most measures (except RT), confirming minimal practice effects and providing a robust baseline for evaluating microgravity’s impact on the taikonauts. All ground tests, for both groups, were conducted using identical devices and software. The layout of the table and tablet was kept consistent across sessions. The only difference was that, in ground tests, participants sat in a chair approximately 90 cm above the ground without foot straps and did not need to hold the edge of the tabletop with their left hand for stability.

Model simulation

We simulated the peak speed, peak acceleration, and their corresponding times for reaching movements in different directions. We employed the classical two-joint arm model to demonstrate the effect of biomechanics (Li and Todorov, 2004; Figure 1B). This was combined with a movement utility model to estimate planned movement time (Shadmehr et al., 2016; Figure 1—figure supplement 1A, B) and a forward optimal controller to simulate reaching movements (Todorov, 2005; Figure 1C, F, Figure 1—figure supplement 1C, D). Our simulation is briefly outlined below in steps.

We used a two-joint planar arm model to estimate the effective mass of the hand moving in different directions (Figure 1B). The parameters of the upper arm (length $l_{1}$ , mass $m_{1}$ , center of mass $l c_{1}$ , inertia $I_{1}$ ) and the lower arm (length $l_{2}$ , mass $m_{2}$ , center of mass $l c_{2}$ , inertia $I_{2}$ ) are based on values from Shadmehr et al., 2016:

l_{1} = 0.33, l_{2} = 0.43 m

m_{1} = 1.93, m_{2} = 1.52 k g

l_{c 1} = l_{1} / 2 l_{c 2} = 2 l_{2} / 3 m

I_{1} = 0.014, I_{2} = 0.019 k g m^{2}

The elbow and shoulder angles are denoted by $θ_{e}$ and $θ_{s}$ , respectively. The initial arm configuration is set by $θ_{s} = 0.785$ rad, and $θ_{e} = 1.571$ rad, following the experimental conditions from Gordon et al., 1994. With a particular joint angle configuration, the endpoint hand position is:

x = [l_{1} c o s θ_{s} + l_{2} c o s (θ_{s} + θ_{e}); l_{1} s i n θ_{s} + l_{2} s i n (θ_{s} + θ_{e})] .

The arm’s inertia matrix is:

I (θ_{s}, θ_{e}) = [I_{11}, I_{12}; I_{21}, I_{22}],

where

I_{11} = a_{3} + a_{1} l_{1}^{2} + a_{4} + 2 a_{2} l_{1} \cos θ_{e}, I_{12} = I_{21} = a_{2} + l_{1} \cos θ_{e} + a_{4}, I_{22} = a_{4},

with

a_{1} = m_{2}, a_{2} = m_{2} l_{c 2}, a_{3} = m_{1} l_{c 1}^{2} + i_{1}, a_{4} = m_{2} l_{c 2}^{2} + i_{2}

The joint torques at the shoulder and elbow are represented as:

τ = I (θ) \ddot{θ},

and the forces at the hand are:

f = M (θ) \ddot{x} .

We use the Jacobian matrix $Λ = \frac{d x}{d θ}$ to relate force and torque, based on the virtual work principle:

τ = Λ^{T} f,

\dot{x} = Λ \dot{θ},

\ddot{x} = \dot{Λ} \dot{θ} + Λ \ddot{θ} .

Put Equation 3 into Equation 5:

f = Λ^{- 1^{T}} I (θ) \ddot{θ},

then combine Equations 7 and 8:

f = Λ^{- 1^{T}} I (θ) Λ^{- 1} (\ddot{x} - \dot{Λ} \dot{θ}) .

Since the hand speed at the beginning of the movement is zero, according to Equations 4 and 9, the mass matrix of hand $M (θ)$ is:

M (θ) = Λ^{- 1^{T}} I (θ) Λ^{- 1} .

The mass matrix $M (θ)$ is a 2x2 matrix, and the effective mass $m (θ)$ is determined by the length of the force vector when $M (θ)$ is subjected to a unit acceleration.

In Figure 1—figure supplement 1A, the solid curve illustrates the effective mass across various directions under normal gravity, while the dashed curve represents the effective mass with a hypothetical 30% ( $m_{0 g} (θ) = 0.3 m (θ)$ ) mass underestimation in microgravity. The colored lines indicate the effective mass amplitudes for the three target directions used in our experiment: 45°, 90°, and 135°.

Movement time determined by optimal utility

Following Shadmehr et al., 2016, the utility of a reaching action is given by:

J_{u t} = \frac{α - a m T - b m d^{i} / T^{2}}{1 + γ T}

where $α$ is the reward, m is mass, d is the distance to be moved, T is the planned duration of the movement, and a, b, and i are scaling factors of the effort cost. $γ$ represents a temporal discounting factor. The optimal movement time ( $T_{o p t}$ ) is the time that maximizes the utility $J_{u t}$ :

T_{o p t} = a r g_{T} m a x (J_{u t}) .

The parameter values used in the simulation of Figure 1—figure supplement 1B were taken from Shadmehr et al., 2016: $α = 1000$ , $a = 15$ , $b = 100$ , $i = 1.1$ , $γ = 1$ . The movement distance $d$ is set as 0.12 m, the target distance in the experiment. Figure 1—figure supplement 1B illustrates how the utility curves change with planned movement time in three experimental directions (45°, 90°, and 135°) under normal gravity, with colored dots indicating the optimal MDs.

Movement kinematics simulated by optimal control theory

After setting the movement time $T_{o p t}$ , the position, speed, and acceleration over time can be simulated based on an optimal control model (Todorov, 2005). The arm’s control operates as a second-order low-pass filter:

u_{t} = τ_{1} τ_{2} {\ddot{f}}_{t} + (τ_{1} - τ_{2}) {\dot{f}}_{t} + f_{t},

with the time constants $τ_{1} = τ_{2} = 40 m s$ .

The system state vector is $x = [p (t); \dot{p} (t); f (t); g (t); p^{*}]$ , where p is the position of the hand, f is the force acting on the hand, and g is an auxiliary state variable, p* is the target position. The control law is given by:

u_{t} = τ_{1} {\dot{g}}_{t} + g_{t}, g_{t} = τ_{2} {\dot{f}}_{t} + f_{t} .

The system dynamics are:

x_{t + 1} = A x_{t} + B (u_{t} + ξ_{t}), y_{t} = H x_{t} + ω_{t}

With the sensory feedback carrying information about position, velocity, and force:

y_{t} = [p (t); \dot{p} (t); f (t)] + ω_{t} .

$ξ_{t}$ is signal-dependent noise $ξ_{k} \sim N (0, {(σ_{c} u_{t})}^{2})$ . $ω_{t}$ is sensory noise $ω_{k} \sim N (0, Ω^{ω})$ , where

Ω^{ω} = d i a g (σ_{p}^{2}, σ_{v}^{2}, σ_{f}^{2})

We define the sampling interval $∆ t$ , and the number of steps is $N = T_{o p t} / Δ t$ , the discrete-time dynamics of the above system are:

p (t + Δ) = p (t) + \dot{p} (t) Δ

\dot{p} (t + Δ) = \dot{p} (t) + f (t) Δ / m

f (t + Δ) = f (t) (1 - \frac{Δ}{τ_{2}}) + g (t) Δ / τ_{2}

g (t + Δ) = g (t) (1 - \frac{Δ}{τ_{1}}) + u (t) (1 + σ_{c} ε_{t}) \frac{Δ}{τ_{1}}

Which is transformed into the matrices as:

A = [\begin{matrix} 1 & Δ & 0 & 0 & 0 \\ 0 & 1 & Δ / m & 0 & 0 \\ 0 & 0 & 1 & Δ / τ_{2} & 0 \\ 0 & 0 & 0 & 1 - Δ / τ_{1} & 0 \\ 0 & 0 & 0 & 0 & 1 \end{matrix}]

H = [\begin{matrix} 1 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & 0 \end{matrix}]

B = [0; 0; 0; Δ / τ_{1}; 0] .

The controller seeks to minimize the cost function J by optimizing the command u:

J = \sum_{t} (x_{t}^{T} Q x_{t} + u_{t}^{T} R u_{t}),

R = w_{u}, Q = p p^{T} + v v^{T} + f f^{T}, p = [w_{p}; 0; 0; 0; - w_{p}], v = [0; w_{v}; 0; 0; 0], f = [0; 0; w_{f}; 0; 0]

Here, we used the penalty values from Todorov, 2005 for each state and control variable: $w_{u} = 0.00001$ , $w_{p} = 2$ , $w_{v} = 0.2$ , $w_{f} = 0.02$ , $σ_{c} = 0.5$ , $Ω^{ω} = {(0.5 d i a g [0.02 m; \frac{0.2 m}{s}; 1 N])}^{2}$ . We first estimated the time-varying gains { $L_{k}$ , $K_{k}$ } correspond to the feedforward mapping and the feedback correction gain, respectively. The control law can be expressed as: $u_{k} = u_{k}^{f f} + u_{k}^{f b} = - L_{k} x_{k}^{n o m} - K_{k} ({\hat{x}}_{k} - x_{k}^{n o m})$ , where $u_{k}$ is the control input, $x_{k}^{n o m}$ is the nominal planned state, ${\hat{x}}_{k}$ is the estimated state, $L_{k}$ is the feedforward (nominal) control associated with the planned trajectory, and $K_{k}$ is the time-varying feedback gain that corrects deviations from the plan. To define the motor plan for comparison with behavior, we then simulate the deterministic open-loop trajectory by turning off noise and disabling feedback corrections, i.e., $u_{k} = u_{k}^{f f} = - L_{k} x_{k}^{n o m}$ . We then obtain the peak speed, peak acceleration, and their respective times.

In summary, we apply the movement utility theory to calculate the optimal movement time for a simplified two-joint arm moving in a 2D plane. This movement time is primarily determined by the effective mass and temporal discounting of reward. Next, we employ the optimal control theory to model the kinematics of hand movement, including its position, speed, and acceleration. This approach also enables us to estimate the peak speed and acceleration, as well as their respective times, for reaches performed toward different target directions. To simulate the effect of microgravity, we introduce a mass underestimation factor to account for the kinematic changes associated with this perceptual error.

Data analysis

Kinematic analysis

A Butterworth low-pass filter with a 20 Hz cutoff frequency was applied to the positional data. Movement distance at each time point was computed as the Euclidean distance between the finger’s current position and the center of the start circle. Speed was estimated by performing linear regression over a moving window of five consecutive positional data points, and acceleration was calculated similarly from the speed data. Movement onset was defined as the first time point at which acceleration exceeded 50 cm/s², while movement offset was the first time point at which acceleration fell below 50 cm/s². RT was defined as the interval between target onset and movement onset, and MD as the interval between movement onset and offset. Movement endpoint error was computed as the Euclidean distance between the target and the finger’s position at movement offset. Peak speed and peak acceleration were identified as the maximum speed and acceleration occurring between movement onset and offset, respectively. The times of peak speed and peak acceleration were measured as their respective time intervals from movement onset, and their relative times were calculated by dividing each by the MD.

Submovement extraction

To separate the primary submovement from the secondary corrective submovement, the speed profile of each reaching movement was decomposed into a sum of submovements. Following the optimization algorithm described by Rohrer and Hogan, 2006, the speed profile was fitted as a sum of support-bounded lognormal (LGNB) curves. Each LGNB curve is defined as:

B (t) = \frac{(T_{1} - T_{0}) D}{σ \sqrt{2 π} (t - T_{0}) (T_{1} - t)} e x p {(\frac{- 1}{2 σ^{2}}) {[l n (\frac{t - T_{0}}{T_{1} - t}) - μ]}^{2}},

where T₀ and T₁ represent the start and end times of the movement, D is a scaling parameter, and the parameters µ and σ determine the skewness and kurtosis of the underlying lognormal function. These five parameters enable each LGNB submovement to vary across a wide range of possible shapes.

In theory, the number of submovements could be treated as a free parameter. However, to avoid overfitting, we limited each movement to two submovements, given the brevity of the movement (average duration approx. 350 ms). For each trial, we initially fit the data using both one and two LGNB curves and then determined the optimal number of submovements using the ‘reedy’ method from the previous study (Rohrer and Hogan, 2006). Briefly, the fitting error for each movement with one and two LGNB curves was calculated as:

ε = \frac{\int | F (t) - G (t) | d t}{\int | G (t) | d t},

where G(t) is the actual movement speed profile, and F(t) is the fitted speed profile (the sum of the submovements). If the fitting error for a single submovement was below a 2% threshold, or if adding a second submovement reduced the fitting error by less than 2%, a single submovement was considered adequate. Otherwise, two submovements were included. Trials with large fitting errors (>10%) were discarded, accounting for 3.60% and 4.83% of trials in the experimental and control groups, respectively. The percentage of trials with two submovements was calculated for each movement direction and phase. The time difference between the peaks of the two submovements was also computed, and set to zero if only one submovement was detected. Data fitting was performed using Matlab’s fmincon function (MathWorks, Natick, MA, USA), and the robustness of fitting was confirmed by testing 20 random sets of initial parameters.

Statistical analysis

First, invalid trials were excluded from further analysis, with three possible causes: early initiation (RT <150 ms, indicating guessing of the target direction), affecting 0.74% and 0.70% of trials in the experimental and control groups, respectively; late initiation (RT >400 ms, indicating inattention), affecting 1.13% and 0%; and measurement failures, affecting 1.16% and 0.70%. The proportion of invalid trials did not differ significantly across testing phases (Friedman’s tests, all p > 0.1).

Most dependent measures—including MD, peak acceleration, peak speed, time to peak acceleration/speed, relative time to peak acceleration/speed, the percentage of submovements, and IPIs—were analyzed using two-way repeated-measures ANOVAs with a 3 (phase) × 3 (movement direction) design. RT was analyzed using three-way repeated-measures ANOVAs with a 3 (phase) × 3 (movement direction) × 2 (beep/no beep) design. Greenhouse–Geisser corrections were applied when sphericity was violated (Kirk, 1968). Normality was assessed with Shapiro–Wilk tests, and if more than 10% of tests for a dependent variable violated normality, the Aligned Rank Transform (ART) procedure was applied before conducting repeated-measures ANOVA. Tukey’s HSD tests were used for pairwise comparisons.

To examine the relationship between submovement and MD, an LMM was used to analyze the association between changes in the IPI of submovements (ΔIPI) and changes in movement duration (ΔMD). The dependent variable $y_{∆ M D}$ represents the mean change in MD across two adjacent phases (pre-to-in and post-to-in transitions) for each target direction. Target direction (45°, 90°, and 135°) and phase transitions (pre-to-in and post-to-in) were treated as within-subject factors. The predictors included three fixed factors: ΔIPI, phase transition ( $x_{p h a}$ ), and direction ( $x_{d i r}$ ). An intercept term ( $u_{p a r}$ ) was included as a random factor to account for individual differences in MD. The resulting LMM is as follows:

y_{Δ M D} = β_{0} + β_{1} x_{Δ I P I} + β_{2} x_{p h a} + β_{3} x_{d i r} + Z u_{p a r} + ε,

where $β_{0}$ denotes the overall intercept and $ε$ denotes a Gaussian residual error.

All statistical analyses were performed using MATLAB’s Statistics and Machine Learning Toolbox. The significance level was set at α = 0.05, and all tests were two-tailed unless otherwise specified.

For the directional dependent effects (Figure 6), we applied repeated measures correlation to quantify within-subject associations (Bakdash and Marusich, 2017). Each participant contributed three repeated measurements under different directions. The predictor variable was identical across conditions, and the outcome variables were the measured responses of interest. Repeated measures correlations were computed separately for each of the four metrics, controlling for between-subject differences by including subject-specific intercepts and estimating a single common slope across participants.

Appendix 1

Cost scaling alone cannot explain earlier timing with lower peaks

To test whether the kinematic changes observed in microgravity—reduced peak velocity and acceleration together with earlier peak times—can be explained solely by changes in optimal control strategy (i.e., cost-function rescaling) without altering the estimated mass.

We used the same LQG framework as in the main text with cost function:

J = \sum_{t} (x_{t}^{T} Q x_{t} + u_{t}^{T} R u_{t}),

planning a primary movement with fixed nominal distance and duration. The simplified arm model used to derive predictions for direction effects is identical to that in the main text; here we hold the estimated mass constant and vary only the control costs.

Following Crevecoeur et al., 2010, we implemented a scaling parameter α by:

Q^{'} = \frac{Q}{α}, R^{'} = α R,

which penalizes control effort more strongly as α increases (more ‘conservative’ strategy), and less strongly as α decreases.

We simulated three movement directions (45°, 90°, and 135°) with alpha = 3.0 (increased effort penalty) and alpha = 0.3 (decreased effort penalty). With alpha = 3.0, peak velocity and acceleration decreased, but peak times were delayed (Figure 1—figure supplement 2); with α = 0.3, peak amplitudes increased and peak times occurred earlier (Figure 1—figure supplement 3). Across directions, no alpha produced the simultaneous combination of lower peak values and earlier peak times that we observed in microgravity. This pattern also mirrors the trade-off in Crevecoeur et al., 2010, Figure 7B, where lowering peak velocity via cost-function scaling is consistently accompanied by later peaks, not earlier ones. We therefore conclude that changes in optimal-control costs alone (via Q/R rescaling) cannot account for the microgravity pattern of reduced peak velocity/acceleration together with earlier peak times.

Appendix 2

Coriolis and centripetal torques have minimal impact

To evaluate the potential contribution of Coriolis and centripetal torques, we examined how adding the Coriolis and centripetal torque term to the optimal control simulations affects the outcomes. Using the iLQR (iterative LQR) method of Li and Todorov, 2004, we generated OFC-based hand-movement trajectories (Figures R4 and R5) and computed the contributions of each torque component separately (Figure R6). Key methods and results are summarized below.

Following Li and Todorov, 2004, the forward dynamics of joint angles are given by:

\ddot{θ} = M {(θ)}^{- 1} (τ - C (θ, \dot{θ})),

where

M = (\begin{matrix} a_{1} + 2 a_{1} \cos θ_{2} & a_{3} + a_{2} \cos θ_{2} \\ a_{3} + a_{2} \cos θ_{2} & a_{3} \end{matrix}), C = (\begin{matrix} - {\dot{θ}}_{2} (2 {\dot{θ}}_{1} + {\dot{θ}}_{2}) \\ {\dot{θ}}_{1}^{2} \end{matrix}) a_{2} \sin θ_{2},

with parameter definitions:

a_{1} = I_{1} + I_{2} + m_{2} l_{1}^{2}, a_{2} = m_{2} l_{1} s_{2}, a_{3} = I_{2}

The system is expressed in state-space form:

\dot{x} = F (x) + G (x) u,

where

x = (θ_{1} θ_{2} {\dot{θ}}_{1} {\dot{θ}}_{2})^{T}, u = (τ = τ_{1} τ_{2})^{T}

The cost function is defined as:

J_{0} = \frac{1}{2} {(θ (T) - θ^{*})}^{T} (θ (T) - θ^{*}) + \frac{1}{2} \int_{0}^{T} γ τ^{T} τ d t .

All mechanical parameters and muscle time constants were kept consistent with those used in the main text. Because the dynamics are nonlinear, we solved the OFC problem using the iLQR/iLQG trajectory optimization scheme (Li and Todorov, 2004), which fits a sequence of local LQR subproblems (backward Riccati pass and forward rollout) to obtain a locally optimal time-varying feedback controller for the nonlinear system.

We extracted the hand endpoint position as:

x (θ) = [\begin{matrix} l_{1} c o s θ_{1} + l_{2} c o s (θ_{1} + θ_{2}) \\ l_{1} s i n θ_{1} + l_{2} s i n (θ_{1} + θ_{2}) \end{matrix}] .

Hand speed and acceleration were computed (Figure 1—figure supplement 4, left panels, solid lines). Using the same utility model to determine movement times, we also simulated hand position, speed, and acceleration under microgravity (dashed lines). The right panels replicate results from the simplified model. Overall, the model with Coriolis and centripetal torques produces similar trajectories to the original, simplified model.

Figure 1—figure supplement 4 compares simulation results from the full model (with Coriolis and centripetal torques, left panels) and the simplified model (right panels). The top row shows hand position profiles, with speed and acceleration shown below. The three colors represent different movement directions (dark red: 45°, red: 90°, yellow: 135°). Dashed lines show simulations under microgravity, assuming mass underestimation. In the full model, a two-link arm was used with both inertial and Coriolis and centripetal torque terms.

These results support our original simplification: excluding Coriolis and centripetal torques does not affect the model’s core predictions about mass underestimation and its effects on motor behavior.

To further evaluate Coriolis and centripetal torque contributions, we followed Hollerbach and Flash, 1982 by decomposing net torque τ into inertial and CC components:

τ = I (θ) \ddot{θ} + C (θ, \dot{θ}) \dot{θ}, τ_{i n e} = I (θ) \ddot{θ}, τ_{c o r} = C (θ, \dot{θ}) \dot{θ} .

As shown in Figure 1—figure supplement 5, the Coriolis and centripetal term (dashed red) contributes only modestly to the net torque, while the inertial term (dotted red) dominates. This differs from Hollerbach and Flash, likely because their movements involved larger joint excursions and higher peak speeds (≈30–90° vs. <30° here) and larger peak torques (~15 vs. ~4 N m here). Given the small 12 cm reaches in our paradigm, Coriolis and centripetal torques remain comparatively minor across all three directions, and they had only a negligible effect on overall kinematics, as shown in Figure R4. We, therefore, conclude that inertial torques—modulated by mass underestimation—are the primary drivers of movement slowing in microgravity. Misestimation of Coriolis and centripetal torques cannot explain our empirical findings, since Coriolis and centripetal torques play only a minor role in this context.

Data availability

The control-group data are openly available at OSF (https://osf.io/3ac5b), and the analysis code (MATLAB) is available at GitHub (https://github.com/ZhaoranZhang/Aiming_data_analysis, copy archived at Zhang, 2026). Group-level data for the astronauts, underlying all figures and statistical tests, are reported within the main text and Supplementary file 2. Individual-level data for the astronauts cannot be made publicly available, in accordance with the information security protocols of the China Astronaut Research and Training Center. In addition, the astronaut cohort is small and its members are publicly identifiable; individual-level data, therefore, cannot be effectively de-identified without an unacceptable risk of re-identification. For these reasons, a de-identified individual-level dataset cannot be released either. Researchers wishing to access these restricted data should submit a written request and a brief project proposal describing the intended use to the corresponding author (Kunlin Wei, wei.kunlin@pku.edu.cn). Requests are assessed by the Ethics Committee of the China Astronaut Research and Training Center, normally within 4–6 weeks. Access is at the Center's discretion and may be subject to a data-use agreement. Commercial use of the data is not permitted.

The following data sets were generated

1. Zhang Z
(2026) Open Science Framework
ID 3ac5b. Space_flight_aiming.

https://osf.io/3ac5b

References

(2018) Contact mechanics between the human finger and a touchscreen under electroadhesion
PNAS 115:12668–12673.

https://doi.org/10.1073/pnas.1811750115
- PubMed
- Google Scholar
1. Bakdash JZ
2. Marusich LR
(2017) Repeated measures correlation
Frontiers in Psychology 8:456.

https://doi.org/10.3389/fpsyg.2017.00456
- PubMed
- Google Scholar
(1997)
Pointing arm movements in short- and long-term spaceflights

Aviation, Space, and Environmental Medicine 68:781–787.
- PubMed
- Google Scholar
1. Berniker M
2. Kording K
(2008) Estimating the sources of motor errors for adaptation and generalization
Nature Neuroscience 11:1454–1461.

https://doi.org/10.1038/nn.2229
- PubMed
- Google Scholar
(1992)
Accuracy of aimed arm movements in changed gravity

Aviation, Space, and Environmental Medicine 63:994–998.
- PubMed
- Google Scholar
(1996a)
Performance of a simple aiming task in hypergravity: I. overall accuracy

Aviation, Space, and Environmental Medicine 67:127–132.
- PubMed
- Google Scholar
(1996b)
Performance of a simple aiming task in hypergravity: II. detailed response characteristics

Aviation, Space, and Environmental Medicine 67:133–138.
- PubMed
- Google Scholar
1. Bock O
(1998) Problems of sensorimotor coordination in weightlessness
Brain Research. Brain Research Reviews 28:155–160.

https://doi.org/10.1016/s0165-0173(98)00035-6
- PubMed
- Google Scholar
(2001)
Human sensorimotor coordination during spaceflight: an analysis of pointing and tracking responses during the “Neurolab” Space Shuttle mission

Aviation, Space, and Environmental Medicine 72:877–883.
- PubMed
- Google Scholar
Book
1. Bock O
2. Fowler B
3. Jüngling S
4. Comfort D
(2003)
Visual-motor coordination during spaceflight

In: Homick J, Buckey J, editors. The Neurolab Space Mission: Neuroscience Research in Space. National Aeronautics and Space Administration. pp. 83–89.
- Google Scholar
(2010) Cognitive demand of human sensorimotor performance during an extended space mission: a dual-task study
Aviation, Space, and Environmental Medicine 81:819–824.

https://doi.org/10.3357/asem.2608.2010
- PubMed
- Google Scholar
Book
(2025) Microgravity-Induced Changes in Skeletal Muscle and Possible Countermeasures: What We Can Learn from Bed Rest and Human Space Studies
Experimental Physiology.

https://doi.org/10.1113/EP092345
- Google Scholar
Book
1. Brooks VB
(1986)
The Neural Basis of Motor Control

Oxford: Oxford University Press.
- Google Scholar
1. Chua R
2. Elliott D
(1993) Visual regulation of manual aiming
Human Movement Science 12:365–401.

https://doi.org/10.1016/0167-9457(93)90026-L
- Google Scholar
1. Craik KJW
(1947) Theory of the human operator in control systems; the operator as an engineering system
The British Journal of Psychology 38:56–61.

https://doi.org/10.1111/j.2044-8295.1947.tb01141.x
- PubMed
- Google Scholar
(2010) Movement stability under uncertain internal models of dynamics
Journal of Neurophysiology 104:1301–1313.

https://doi.org/10.1152/jn.00315.2010
- PubMed
- Google Scholar
(2014) Gravity-dependent estimates of object mass underlie the generation of motor commands for horizontal limb movements
Journal of Neurophysiology 112:384–392.

https://doi.org/10.1152/jn.00061.2014
- PubMed
- Google Scholar
(2013) The temporal evolution of feedback gains rapidly update to task demands
The Journal of Neuroscience 33:10898–10909.

https://doi.org/10.1523/JNEUROSCI.5669-12.2013
- PubMed
- Google Scholar
(1998) Cognitive performance aboard the life and microgravity spacelab
Acta Astronautica 43:193–210.

https://doi.org/10.1016/s0094-5765(98)00154-4
- PubMed
- Google Scholar
1. Elliott D
2. Hansen S
3. Grierson LEM
4. Lyons J
5. Bennett SJ
6. Hayes SJ
(2010) Goal-directed aiming: two components but multiple processes
Psychological Bulletin 136:1023–1044.

https://doi.org/10.1037/a0020958
- PubMed
- Google Scholar
1. Elliott D
2. Lyons J
3. Hayes SJ
4. Burkitt JJ
5. Roberts JW
6. Grierson LEM
7. Hansen S
8. Bennett SJ
(2017) The multiple process model of goal-directed reaching revisited
Neuroscience and Biobehavioral Reviews 72:95–110.

https://doi.org/10.1016/j.neubiorev.2016.11.016
- PubMed
- Google Scholar
1. Fercho K
2. Baugh LA
(2014) It’s too quick to blame myself-the effects of fast and slow rates of change on credit assignment during object lifting
Frontiers in Human Neuroscience 8:554.

https://doi.org/10.3389/fnhum.2014.00554
- PubMed
- Google Scholar
(1993) Gravitoinertial force level influences arm movement control
Journal of Neurophysiology 69:504–511.

https://doi.org/10.1152/jn.1993.69.2.504
- PubMed
- Google Scholar
1. Flanagan JR
2. Wing AM
(1997) The role of internal models in motion planning and control: evidence from grip force adjustments during movements of hand-held loads
The Journal of Neuroscience 17:1519–1528.

https://doi.org/10.1523/JNEUROSCI.17-04-01519.1997
- PubMed
- Google Scholar
1. Flash T
2. Hogan N
(1985) The coordination of arm movements: an experimentally confirmed mathematical model
The Journal of Neuroscience 5:1688–1703.

https://doi.org/10.1523/JNEUROSCI.05-07-01688.1985
- PubMed
- Google Scholar
(2008) Perceptual-motor performance and associated kinematics in space
Human Factors 50:879–892.

https://doi.org/10.1518/001872008X374965
- PubMed
- Google Scholar
(2016) Direction-dependent arm kinematics reveal optimal integration of gravity cues
eLife 5:e16394.

https://doi.org/10.7554/eLife.16394
- PubMed
- Google Scholar
1. Gordon J
2. Ghilardi MF
3. Cooper SE
4. Ghez C
(1994) Accuracy of planar reaching movements. II. Systematic extent errors resulting from inertial anisotropy
Experimental Brain Research 99:112–130.

https://doi.org/10.1007/BF00241416
- PubMed
- Google Scholar
1. Heuer H
2. Manzey D
3. Lorenz B
4. Sangals J
(2003) Impairments of manual tracking performance during spaceflight are associated with specific effects of microgravity on visuomotor transformations
Ergonomics 46:920–934.

https://doi.org/10.1080/0014013031000107559
- PubMed
- Google Scholar
1. Hollerbach JM
2. Flash T
(1982) Dynamic interactions between limb segments during planar arm movement
Biological Cybernetics 44:67–77.

https://doi.org/10.1007/BF00353957
- Google Scholar
1. Johansson R
2. Edin B
(1993)
Predictive feed-forward sensory control during grasping and manipulation in man

Biomedical Research-Tokyo 14:95–106.
- Google Scholar
Book
1. Kanas N
2. Manzey D
(2008) Basic issues of human adaptation to space flight
In: Kanas N, editors. Space Psychology and Psychiatry. Dordrecht: Springer. pp. 15–48.

https://doi.org/10.1007/978-1-4020-6770-9_2
- Google Scholar
1. Kawato M
(1999) Internal models for motor control and trajectory planning
Current Opinion in Neurobiology 9:718–727.

https://doi.org/10.1016/s0959-4388(99)00028-8
- PubMed
- Google Scholar
Book
1. Kirk R
(1968)
Experimental Design: Procedures for the Behavioral Sciences

Belmont, California, USA: Brooks/Cole.
- Google Scholar
(2008) Reach adaptation: what determines whether we learn an internal model of the tool or adapt the model of our arm?
Journal of Neurophysiology 100:1455–1464.

https://doi.org/10.1152/jn.90334.2008
- PubMed
- Google Scholar
1. Kong G
2. Zhou Z
3. Wang Q
4. Kording K
5. Wei K
(2017) Credit assignment between body and object probed by an object transportation task
Scientific Reports 7:13415.

https://doi.org/10.1038/s41598-017-13889-w
- PubMed
- Google Scholar
(2000) Learning of visuomotor transformations for vectorial planning of reaching trajectories
The Journal of Neuroscience 20:8916–8924.

https://doi.org/10.1523/JNEUROSCI.20-23-08916.2000
- PubMed
- Google Scholar
1. Krakauer JW
2. Hadjiosif AM
3. Xu J
4. Wong AL
5. Haith AM
(2019) Motor Learning
Comprehensive Physiology 9:613–663.

https://doi.org/10.1002/cphy.c170043
- PubMed
- Google Scholar
Report
(1977)
Task and work performance on Skylab missions 2, 3, and 4: Time and motion study: Experiment M151

NASA. JSC Biomed. Results from Skylab.
- Google Scholar
(2008) Long-latency reflexes of the human arm reflect an internal model of limb dynamics
Current Biology 18:449–453.

https://doi.org/10.1016/j.cub.2008.02.053
- PubMed
- Google Scholar
1. Lackner JR
2. DiZio P
(1992) Gravitoinertial force level affects the appreciation of limb position during muscle vibration
Brain Research 592:175–180.

https://doi.org/10.1016/0006-8993(92)91673-3
- PubMed
- Google Scholar
1. Lackner JR
2. Dizio P
(1994) Rapid adaptation to Coriolis force perturbations of arm trajectory
Journal of Neurophysiology 72:299–313.

https://doi.org/10.1152/jn.1994.72.1.299
- PubMed
- Google Scholar
(2001) Effect of long-duration spaceflight on postural control during self-generated perturbations
Journal of Applied Physiology 90:997–1006.

https://doi.org/10.1152/jappl.2001.90.3.997
- PubMed
- Google Scholar
Conference
1. Li W
2. Todorov E
(2004)
Iterative linear quadratic regulator design for nonlinear biological movement systems

First International Conference on Informatics in.
- Google Scholar
(1998) Mental performance in extreme environments: results from a performance monitoring study during a 438-day spaceflight
Ergonomics 41:537–559.

https://doi.org/10.1080/001401398186991
- PubMed
- Google Scholar
(1976) Stretch reflex and servo action in a variety of human muscles
The Journal of Physiology 259:531–560.

https://doi.org/10.1113/jphysiol.1976.sp011481
- Google Scholar
(2001) Does the brain model Newton’s laws?
Nature Neuroscience 4:693–694.

https://doi.org/10.1038/89477
- PubMed
- Google Scholar
(2002) Slowing of human arm movements during weightlessness: the role of vision
European Journal of Applied Physiology 87:576–583.

https://doi.org/10.1007/s00421-002-0684-3
- PubMed
- Google Scholar
1. Meyer DE
2. Abrams RA
3. Kornblum S
4. Wright CE
5. Smith JE
(1988) Optimality in human motor performance: ideal control of rapid aimed movements
Psychological Review 95:340–370.

https://doi.org/10.1037/0033-295x.95.3.340
- PubMed
- Google Scholar
Book
(2018) Speed—accuracy tradeoffs in aimed movements: toward a theory of rapid voluntary action
In: Meyer D, Keith Smith JE, editors. Attention and Performance Xiii. Lawrence Erlbaum Associates, Inc. pp. 1–54.

https://doi.org/10.4324/9780203772010-6
- Google Scholar
(1985) Neural, mechanical, and geometric factors subserving arm posture in humans
The Journal of Neuroscience 5:2732–2743.

https://doi.org/10.1523/JNEUROSCI.05-10-02732.1985
- PubMed
- Google Scholar
1. Newman DJ
2. Lathan CE
(1999) Memory processes and motor control in extreme environments
IEEE Transactions on Systems, Man and Cybernetics. Part C, Applications and Reviews: A Publication of the IEEE Systems, Man, and Cybernetics Society 29:387–394.

https://doi.org/10.1109/5326.777074
- PubMed
- Google Scholar
(1992) Recovery of postural equilibrium control following spaceflight
Annals of the New York Academy of Sciences 656:747–754.

https://doi.org/10.1111/j.1749-6632.1992.tb25253.x
- PubMed
- Google Scholar
(1998) Hand trajectories of vertical arm movements in one-G and zero-G environments. Evidence for a central representation of gravitational force
Experimental Brain Research 120:496–502.

https://doi.org/10.1007/s002210050423
- PubMed
- Google Scholar
(2005) Kinematic and dynamic processes for the control of pointing movements in humans revealed by short-term exposure to microgravity
Neuroscience 135:371–383.

https://doi.org/10.1016/j.neuroscience.2005.06.063
- PubMed
- Google Scholar
1. Proske U
2. Gandevia SC
(2012) The proprioceptive senses: their roles in signaling body shape, body position and movement, and muscle force
Physiological Reviews 92:1651–1697.

https://doi.org/10.1152/physrev.00048.2011
- PubMed
- Google Scholar
1. Proske U
2. Weber BM
(2023) Proprioceptive disturbances in weightlessness revisited
NPJ Microgravity 9:64.

https://doi.org/10.1038/s41526-023-00318-8
- PubMed
- Google Scholar
(1998) Posture, locomotion, spatial orientation, and motion sickness as a function of space flight
Brain Research. Brain Research Reviews 28:102–117.

https://doi.org/10.1016/s0165-0173(98)00031-9
- PubMed
- Google Scholar
1. Rohrer B
2. Hogan N
(2006) Avoiding spurious submovement decompositions II: a scattershot algorithm
Biological Cybernetics 94:409–414.

https://doi.org/10.1007/s00422-006-0055-y
- PubMed
- Google Scholar
1. Ross HE
2. Reschke MF
(1982) Mass estimation and discrimination during brief periods of zero gravity
Perception & Psychophysics 31:429–436.

https://doi.org/10.3758/BF03204852
- Google Scholar
1. Sangals J
2. Heuer H
3. Manzey D
4. Lorenz B
(1999) Changed visuomotor transformations during and after prolonged microgravity
Experimental Brain Research 129:378–390.

https://doi.org/10.1007/s002210050906
- PubMed
- Google Scholar
(2023) Effects of exercise countermeasures on multisystem function in long duration spaceflight astronauts
NPJ Microgravity 9:11.

https://doi.org/10.1038/s41526-023-00256-5
- PubMed
- Google Scholar
1. Shadmehr R
2. Mussa-Ivaldi FA
(1994) Adaptive representation of dynamics during learning of a motor task
The Journal of Neuroscience 14:3208–3224.

https://doi.org/10.1523/JNEUROSCI.14-05-03208.1994
- PubMed
- Google Scholar
(2016) A representation of effort in decision-making and motor control
Current Biology 26:1929–1934.

https://doi.org/10.1016/j.cub.2016.05.065
- PubMed
- Google Scholar
1. Smeets JB
2. Brenner E
(1995) Prediction of a moving target’s position in fast goal-directed action
Biological Cybernetics 73:519–528.

https://doi.org/10.1007/BF00199544
- PubMed
- Google Scholar
(2014) Human cognitive performance in spaceflight and analogue environments
Aviation, Space, and Environmental Medicine 85:1033–1048.

https://doi.org/10.3357/ASEM.3961.2014
- PubMed
- Google Scholar
(2021) Movement preparation time determines movement variability
Journal of Neurophysiology 125:2375–2383.

https://doi.org/10.1152/jn.00087.2020
- PubMed
- Google Scholar
1. Tafforin C
2. Thon B
3. Guell A
4. Campan R
(1989)
Astronaut behavior in an orbital flight situation: preliminary ethological observations

Aviation, Space, and Environmental Medicine 60:949–956.
- PubMed
- Google Scholar
1. Tays GD
2. Hupfeld KE
3. McGregor HR
4. Salazar AP
5. De Dios YE
6. Beltran NE
7. Reuter-Lorenz PA
8. Kofman IS
9. Wood SJ
10. Bloomberg JJ
11. Mulavara AP
12. Seidler RD
(2021) The effects of long duration spaceflight on sensorimotor control and cognition
Frontiers in Neural Circuits 15:723504.

https://doi.org/10.3389/fncir.2021.723504
- PubMed
- Google Scholar
1. Tian Y
2. Zhang Z
3. Jiang C
4. Chen D
5. Liu Z
6. Wei M
7. Wang C
8. Wei K
(2024) Stressors affect human motor timing during spaceflight
NPJ Microgravity 10:108.

https://doi.org/10.1038/s41526-024-00439-8
- PubMed
- Google Scholar
1. Todorov E
(2004) Optimality principles in sensorimotor control
Nature Neuroscience 7:907–915.

https://doi.org/10.1038/nn1309
- PubMed
- Google Scholar
1. Todorov E
(2005) Stochastic optimal control and estimation methods adapted to the noise characteristics of the sensorimotor system
Neural Computation 17:1084–1108.

https://doi.org/10.1162/0899766053491887
- PubMed
- Google Scholar
1. Wang J
2. Sainburg RL
(2004) Interlimb transfer of novel inertial dynamics is asymmetrical
Journal of Neurophysiology 92:349–360.

https://doi.org/10.1152/jn.00960.2003
- PubMed
- Google Scholar
Book
(2019)
Disentangling the Enigmatic Slowing Effect of Microgravity on Sensorimotor Performance

HFES-Europe.
- Google Scholar
1. Weber B
2. Stelzer M
(2022) Sensorimotor impairments during spaceflight: Trigger mechanisms and haptic assistance
Frontiers in Neuroergonomics 3:959894.

https://doi.org/10.3389/fnrgo.2022.959894
- PubMed
- Google Scholar
(2013) Sensorimotor recalibration depends on attribution of sensory prediction errors to internal causes
PLOS ONE 8:e54925.

https://doi.org/10.1371/journal.pone.0054925
- PubMed
- Google Scholar
1. Winnard A
2. Scott J
3. Waters N
4. Vance M
5. Caplan N
(2019) Effect of time on human muscle outcomes during simulated microgravity exposure without countermeasures-systematic review
Frontiers in Physiology 10:1046.

https://doi.org/10.3389/fphys.2019.01046
- PubMed
- Google Scholar
(1995) An internal model for sensorimotor integration
Science 269:1880–1882.

https://doi.org/10.1126/science.7569931
- PubMed
- Google Scholar
1. Wolpert DM
2. Miall RC
(1996) Forward models for physiological motor control
Neural Networks 9:1265–1279.

https://doi.org/10.1016/s0893-6080(96)00035-4
- PubMed
- Google Scholar
(1998) Internal models in the cerebellum
Trends in Cognitive Sciences 2:338–347.

https://doi.org/10.1016/S1364-6613(98)01221-2
- Google Scholar
(2011) Principles of sensorimotor learning
Nature Reviews. Neuroscience 12:739–751.

https://doi.org/10.1038/nrn3112
- PubMed
- Google Scholar
1. Woodworth RS
(1899) Accuracy of voluntary movement
The Psychological Review 3:2992.

https://doi.org/10.1037/h0092992
- Google Scholar
Software
1. Zhang Z
(2026) Aiming_data_analysis, version swh:1:rev:72afaf5c1aec0da49ce503e2f3eee40f38e1b98a
Software Heritage.

https://archive.softwareheritage.org/swh:1:dir:be55798c8ca296119b30d5ac698360acc71593c1;origin=https://github.com/ZhaoranZhang/Aiming_data_analysis;visit=swh:1:snp:e80c2eeef4bd643a3ed523ed874d1c79d864eb33;anchor=swh:1:rev:72afaf5c1aec0da49ce503e2f3eee40f38e1b98a

Article and author information

Author details

Zhaoran Zhang
1. School of Psychological and Cognitive Sciences, Peking University, Beijing, China
2. School of Psychology, Shenzhen University, Shenzhen, China
Contribution
Conceptualization, Data curation, Formal analysis, Funding acquisition, Visualization, Methodology, Writing – original draft, Writing – review and editing

Contributed equally with
Yu Tian

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-4192-4088
Yu Tian

National Key Laboratory of Human Factors Engineering, China Astronaut Research and Training Center, Beijing, China

Contribution
Conceptualization, Data curation, Funding acquisition, Methodology

Contributed equally with
Zhaoran Zhang

Competing interests
No competing interests declared
Chunhui Wang

National Key Laboratory of Human Factors Engineering, China Astronaut Research and Training Center, Beijing, China

Contribution
Data curation, Funding acquisition

Competing interests
No competing interests declared
Changhua Jiang

National Key Laboratory of Human Factors Engineering, China Astronaut Research and Training Center, Beijing, China

Contribution
Data curation

Competing interests
No competing interests declared
Bo Wang

National Key Laboratory of Human Factors Engineering, China Astronaut Research and Training Center, Beijing, China

Contribution
Data curation

Competing interests
No competing interests declared
Hongqiang Yu

National Key Laboratory of Human Factors Engineering, China Astronaut Research and Training Center, Beijing, China

Contribution
Data curation

Competing interests
No competing interests declared
Rui Zhao

National Key Laboratory of Human Factors Engineering, China Astronaut Research and Training Center, Beijing, China

Contribution
Data curation

Competing interests
No competing interests declared
Kunlin Wei
1. School of Psychological and Cognitive Sciences, Peking University, Beijing, China
2. Beijing Key Laboratory of Behavior and Mental Health, Beijing, China
Contribution
Conceptualization, Supervision, Funding acquisition, Methodology, Writing – original draft, Writing – review and editing

For correspondence
wei.kunlin@pku.edu.cn

Competing interests
Reviewing editor, eLife

"This ORCID iD identifies the author of this article:" 0000-0001-5098-3808

Funding

STI2030-Major Projects (2021ZD0202600)

Kunlin Wei

Space Medical Experiment Project of China Manned Space Program (HYZHXM03002)

Kunlin Wei

Space Medical Experiment Project of China Manned Space Program (HYZHXMR01001)

Kunlin Wei

National Natural Science Foundation of China (62061136001)

Kunlin Wei

National Natural Science Foundation of China (32071047)

Kunlin Wei

National Natural Science Foundation of China (31871102)

Kunlin Wei

National Natural Science Foundation of China (32300868)

Zhaoran Zhang

Shenzhen University (000001033416)

Zhaoran Zhang

Foundation of National Key Laboratory of Human Factors Engineering (HFNKL2023J02)

Yu Tian

National Key Research and Development Program of China (2023YFF1203905)

Chunhui Wang

The funders had no role in study design, data collection, and interpretation, or the decision to submit the work for publication.

Acknowledgements

We would like to thank Mr. Shaolin Zeng for his help with data acquisition of the control group. This work was supported by STI2030-Major Projects (2021ZD0202600), the Space Medical Experiment Project of China Manned Space Program (HYZHXM03002, HYZHXMR01001), and the National Natural Science Foundation of China (62061136001, 32071047, and 31871102), awarded to KW; the National Natural Science Foundation of China (32300868) and a start-up fund (000001033416) from Shenzhen University awarded to ZZ; the Foundation of National Key Laboratory of Human Factors Engineering (HFNKL2023J02) awarded to YT; and the National Key Research and Development Program of China under Grant (2023YFF1203905) awarded to CW.

Ethics

All participants were provided with written informed consent forms, approved by the Ethics Committee of the China Astronaut Research and Training Center and the Institutional Review Board of Peking University.

Version history

Sent for peer review: May 13, 2025
Preprint posted: May 15, 2025
Reviewed Preprint version 1: July 14, 2025
Reviewed Preprint version 2: January 14, 2026
Reviewed Preprint version 3: June 22, 2026
Version of Record published: July 9, 2026

Cite all versions

You can cite all versions using the DOI https://doi.org/10.7554/eLife.107472. This DOI represents all versions, and will always resolve to the latest one.

Copyright

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.