Model simulations revealing characteristic kinematic features of reaching movements.

A) Speed profile predictions under two competing hypotheses. Under normal gravity, hand speed follows a typical profile (bold apricot lines). According to the body mass underestimation hypothesis (left 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 (right 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°, 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 provided in Supplementary Note 1.

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).

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.

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.

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).

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 1 or 2 in-flight sessions.

Simulation details for the arm model.

A) Effective mass across different movement directions, with three colored lines indicating the specific directions used in the experiment. The solid-line contour denotes the effective mass simulated by default parameters under normal gravity (see Supplemental Text 1), and the dash-line contour denotes a hypothetical 30% underestimation of the effective mass in microgravity B) Utility function values for movements in each of the three directions. C-D) Speed and acceleration profiles for movements in these directions. E-H) Simulated values of peak speed, peak acceleration, peak speed timing, and peak acceleration timing across all movement directions. The solid-line contours denote simulated values by default parameters, and the dash-line contours denotes the values with a 30% mass underestimation.

Movement accuracy quantified by endpoint error.

The average endpoint error for each target was analyzed across experimental phases. Both the experimental group A) and control group B) performed the task with high accuracy despite the stringent timing requirements. The presence of a beep had no effect on endpoint error in either group (experimental group: main effect, F(1,11) = 1.4461, p = 0.254; all interactions, p > 0.2; control group: main effect, F(1,11) = 0.717, p = 0.415; all interactions, p > 0.25). Therefore, data from the two beep conditions were pooled, and a two-way repeated-measures ANOVA was conducted for each group. In the experimental group, endpoint error showed a slight decrease across phases, but the main effect of phase was not statistically significant (F(2,22) = 1.768, p = 0.210), nor was the interaction effect (F(2,22) = 2.386, p = 0.091). However, there was a significant main effect of direction (F(2,22) = 16.020, p < 0.001, partial η2 = 0.593), with post-hoc comparisons revealing that the 45° direction had significantly larger endpoint errors than the 90° and 135° directions (p = 0.002, partial η2 = 0.291; p = 0.001, partial η2 = 0.286, respectively). In the control group, no significant effects of phase or direction were observed, and there was no significant interaction (all p > 0.23).

Results of kinematic analysis of the control group.

The control group exhibited no phase effect for most performance measures, with the exception of reaction time (RT). However, similar effects were also observed in the experimental group. Each measure was analyzed using a 3 (direction) × 3 (phase) two-way repeated-measures ANOVA. A) Reaction Time (RT): Significant main effects of direction (F(2,22) = 21.946, p < 0.001, partial η2 = 0.666) and phase (F(2,22) = 4.281, p = 0.039, partial η2 = 0.280) were found, along with a significant direction-phase interaction (F(2,22) = 7.940, p = 0.001, partial η2 = 0.419). RT was significantly faster in the Middle phase than in the Early phase (p = 0.032, D = 0.285). B) Movement Duration (MD): Significant main effect of direction (F(2,22) = 99.874, p < 0.001, partial η2 = 0.901), with all pairwise comparisons significant at p < 0.001. Neither main effect of phase (F(2,22) = 0.995, p = 0.360) nor interaction effect (F(4,44) = 1.412, p = 0.114) was significant. C) Peak acceleration: Significant main effect of direction (F(2,22) = 55.460, p < 0.001, partial η2 = 0.834), with all pairwise comparisons significant at p < 0.001. A marginal phase effect (F(2,22) = 3.462, p = 0.081) and interaction (F(4,44) = 2.715, p = 0.067) were noted. D) Peak speed: Significant main effect of direction (F(2,22) = 66.071, p < 0.001, partial η2 = 0.857), with all pairwise comparisons significant at p < 0.001. No phase effect (F(2,22) = 2.360, p = 0.148) or interaction (F(4,44) = 1.975, p = 0.145) was observed. E) Peak acceleration time: Significant main effect of direction (F(2,22) = 26.602, p < 0.001, partial η2 = 0.707). No phase effect (F(2,22) = 1.440, p = 0.260) or interaction (F(4,44) = 0.132, p = 0.900) was observed. Pairwise comparisons showed significantly shorter times for 45° compared to 90° (p < 0.001, D = 0.305) and 135° (p < 0.001, D = 0.269). F) Peak speed time: Significant main effect of direction (F(2,22) = 96.575, p < 0.001, partial η2 = 0.898), with all pairwise comparisons significant at p < 0.001. Marginal phase effect (F(2,22) = 2.702, p = 0.098), and no interaction effect (F(4,44) = 0.420, p = 0.710). G) Relative peak acceleration time: Significant main effect of direction (F(2,22) = 8.506, p = 0.003, partial η2 = 0.436), with no phase effect (F(2,22) = 1.143, p = 0.317) or interaction (F(4,44) = 1.002, p = 0.381). Pairwise comparisons indicated that relative peak acceleration time appeared significantly earlier in the 135° direction than in 90° (p = 0.002, D = 0.313) and 45° (p = 0.021, D = 0.256). H) Relative peak speed time: Significant main effect of direction (F(2,22) = 6.779, p = 0.007, partial η2 = 0.381), with a marginal phase effect (F(2,22) = 2.636, p = 0.095) and no interaction (F(4,44) = 1.788, p = 0.175). Pairwise comparisons showed that relative peak speed time was significantly earlier in the 135° direction than in 90° (p = 0.037, D = 0.264) and 45° (p = 0.029, D = 0.308).

Results of submovement analysis of the control group.

All measures of submovements did not show significant phase effect. A) Percentage of trials with two submovements. Only a marginal main effect of direction was detected (F(2,22) = 3.358, p = 0.055, partial η2 = 0.234) without a phase effect (F(2,22) = 2.216, p = 0.155) or interaction (F(4,44) = 1.209, p = 0.320). B) Inter-peak interval (IPI) showed a significant main effect of direction (F(2,22) = 19.277, p < 0.001, partial η2 = 0.637), with no phase effect (F(2,22) = 0.029, p = 0.902) or interaction (F(4,44) = 2.028, p = 0.134). Pairwise comparisons indicated a significantly shorter IPI in the 45° direction compared to the other two directions (45° vs. 90°: p < 0.001, D = 0.269; 45° vs. 135°: p = 0.002, D = 0.305). C) Peak speed of the primary submovement exhibited a significant main effect of direction (F(2,22) = 60.089, p < 0.001, partial η2 = 0.845), without a phase effect (F(2,22) = 2.628, p = 0.120) or interaction (F(4,44) = 0.949, p = 0.420). Pairwise comparisons showed significant differences between all directions (45° vs. 90°: p < 0.001, D = 0.224; 45° vs. 135°: p < 0.001, D = 0.326; 90° vs. 135°: p < 0.001, D = 0.102). D) Peak speed time of the primary submovement revealed a significant main effect of direction (F(2,22) = 46.259, p < 0.001, partial η2 = 0.808), with no phase effect (F(2,22) = 2.335, p = 0.134) or interaction (F(4,44) = 0.425, p = 0.660). Pairwise comparisons showed a significantly smaller peak speed time in the 45° direction compared to the other two directions (45° vs. 90°: p < 0.001, D = 0.260; 45° vs. 135°: p < 0.001, D = 0.311).

Speed-accuracy trade-off as indicated by the relationship between movement duration (MD) and endpoint dispersion, following the classical Fitts’ law.

A) Endpoint dispersion was estimated by constructing a 95% confidence ellipse encompassing the endpoints of all trials. Regardless of movement direction, phase, and beep condition, trials were divided into three equal-sized bins based on duration, labeled as short, medium, and long. For each participant at each phase, trials within each bin were pooled to estimate the ellipse of endpoints. B-C) A positive association between endpoint area and MD was observed, indicating a speed-accuracy trade-off. Both groups exhibited a dependency of the endpoint area on MD. For the experimental group, a 3 (phase) × 3 (MD) two-way repeated-measures ANOVA on endpoint area showed a significant main effect of MD (F(2,22) = 14.125, p = 0.003, partial η2 = 0.407), with only a marginal phase effect and no interaction (phase: F(2,22) = 3.199, p = 0.087; interaction: F(4,44) = 2.106, p = 0.163). Pairwise comparisons indicated that short, medium, and long MD conditions significantly differed from each other (short vs. medium: p = 0.018, D = 0.576; medium vs. long: p = 0.007, D = 0.113; short vs. long: p = 0.004, D = 0.689). The control group yielded similar ANOVA results, with a significant main effect of MD (F(2,22) = 34.058, p < 0.001, partial η2 = 0.756; all post-hoc comparisons were significant at p < 0.007). Additionally, the control group exhibited a main effect of phase (F(2,22) = 9.359, p = 0.003, partial η2 = 0.460) without interaction (F(2,22) = 2.026, p = 0.151, partial η2 = 0.156). Post-hoc comparisons revealed that the early phase had a significantly larger area than the middle and late phases (early vs. middle: p = 0.004, D = 0.263; early vs. late: p = 0.018, D = 0.309). Thus, both groups appear to slightly improve their speed-accuracy trade-off over repeated tests, with no evidence suggesting that spaceflight adversely affects this trade-off.

Speed-accuracy trade-off in action planning, illustrated by the relationship between reaction time (RT) and the variance in initial movement direction.

A) Similar to the speed-accuracy trade-off analysis in Figure S5, all trials were grouped into three equal-sized bins based on RT, regardless of phase, movement direction, or beep condition. The standard deviation of movement direction is shown over time for these bins, using data from the experimental group in the pre-flight phase as an example. The standard deviation decreases over time within a reaching movement, with shorter RTs resulting in greater directional variance, particularly in the early phase of movement. B-C) The standard deviation of initial movement direction as a function of RT, plotted separately for the three phases. A negative trend between standard deviation and RT indicates a speed-accuracy trade-off in action planning. However, phase did not affect the relationship between RT and initial variance. For the experimental group, a 3 (phase) × 3 (RT) two-way repeated measures ANOVA on the standard deviation showed a significant main effect of RT (F(2,22) = 7.542, p = 0.009, partial η2 = 0.407), while neither the main effect of phase nor the interaction was significant (phase: F(2,22) = 3.160, p = 0.083; interaction: F(4,44) = 0.503, p = 0.694). Pairwise comparisons indicated that the initial variance of short-RT trials was significantly higher than that of medium-RT trials (p = 0.004, D = 0.375), while the difference between medium- and long-RT trials was only marginally significant (p = 0.076). The control group showed similar results (RT: F(2,22) = 4.268, p = 0.040, partial η2 = 0.280; phase: F(2,22) = 1.657, p = 0.215; interaction: F(2,22) = 0.007, p = 0.998). Thus, while clear indications of a speed-accuracy trade-off in action planning were observed, there was no evidence that microgravity negatively affected this relationship.

Changes in key performance measures over trials within and between in-flight sessions.

To show a lack of within-session and between-session adaptation during spaceflight, we analyzed changes in peak speed (A), peak speed timing of the primary submovement (B), peak acceleration (C), and peak acceleration timing (D) across the first and second sessions, as well as trials within the second in-flight session. The second session was specifically analyzed because the sensorimotor system undergoes recalibration during early flight and thus the first session is less likely to show good within-session adaptation. The third session had fewer participants (only six taikonauts). Participant S2, who completed only one in-flight session, was excluded from this analysis. For the within-session analysis, we compared the mean values of the first and last five trials for each measure across movement directions. No significant changes were observed in peak speed timing and peak acceleration timing (all p-values > 0.06; panels B and D). Peak speed (panel A) showed a slight and marginally significant decrease in the 45° and 90° directions (45°: from 85.9 ± 3.4 cm/s to 78.5 ± 3.4 cm/s, p = 0.013; 90°: from 71.6 ± 2.1 cm/s to 67.3 ± 1.8 cm/s, p = 0.052), but no significant change in the 135° direction (p = 0.200). Similarly, peak acceleration (panel C) decreased slightly across all directions with marginal significance (45°: from 1033.9 ± 73.4 cm/s2 to 919.9 ± 66.8 cm/s2, p = 0.056; 90°: from 671.8 ± 43.4 cm/s2 to 603.7 ± 33.8 cm/s2, p = 0.097; 135°: from 663.2 ± 46.3 cm/s2 to 586.2 ± 36.2 cm/s2, p = 0.049). For comparisons between the first and second sessions, we applied a two-way repeated measures ANOVA to these dependent variables and found no significant differences between sessions (all p-values > 0.144). These findings suggest that the mass underestimation effects persist over repetitive reaching attempts (120 trials) and practice sessions. Instead of showing between-trial or between-session adaptation, these effects tend to increase slightly across trials.

The schedule of experiment sessions.