Everyday behavior involves performing sequences of actions to achieve a given goal. Sequential motor skills depend on the ability to combine individual actions with short-term goals into a longer sequence consisting of multiple goals (Botvinick and Plaut, 2004). However, the computations that underlie the combination of sequence elements and the neural circuits where this combination occurs remain a topic of debate (Diedrichsen and Kornysheva, 2015; Hikosaka et al., 1999; Wong and Krakauer, 2019).

Behavioral data from various sensorimotor repertoires indicate that the trajectory of sub-movements in a sequence is affected by multiple future goals (Ariani and Diedrichsen, 2019; Diamond et al., 2017; Gottwald et al., 2018; Sheahan et al., 2016; Sosnik et al., 2004; Kashefi et al., 2024). For example, in human speech and bird song, the pronunciation of individual elements is influenced by subsequent elements (Fowler and Saltzman, 1993; Wohlgemuth et al., 2010), a phenomenon known as coarticulation. Ramkumar and colleagues (Ramkumar et al., 2016) employed a computational model, revealing that concatenating multiple sub-movements into larger units, referred to as chunks (Verwey, 2001; Verwey and Abrahamse, 2012), induces coarticulation within the velocity patterns of sub-movements. This coarticulation serves as a natural solution to enhance the efficiency of chunked movements. Collectively, behavioral findings highlight a pattern where sub-movements depend on the whole sequence, or a chunk.

It remains unclear how future goals are integrated in the sensorimotor system. For rapid execution of a sequence, one possible solution is to represent multiple goals within low-level control circuits (Hikosaka et al., 1999; Kadmon Harpaz et al., 2022), enabling the execution of several elements as a single entity, called a ‘motor chunk.’ Note that chunking can also occur at a higher level such as in working memory-guided sequences, which in this case may or may not involve the production of a movement (Mizes et al., 2023; Miller, 1956).

Recent neural recordings in the primary motor cortex (M1) have shown no specific influence of future goals on the population responses governing ongoing action (Yokoi and Diedrichsen, 2019; Zimnik and Churchland, 2021). Specifically, (Zimnik and Churchland, 2021) observed in a two-reach sequence task that, there was no coarticulation in sub-movement kinematics although the execution got faster with practice. Notably, M1 displayed separate phases of execution-related activity for each sub-movement. Using a neural network model, they interpreted that sequence goals could be separated and serially specified to the controller from regions upstream of M1 (Figure 1A). These findings contrast with earlier studies showing coarticulation of sub-movements and whole sequence representations in M1 (Ben-Shaul et al., 2004; Hatsopoulos et al., 2003; Lu and Ashe, 2005). As a result, it has been suggested that coarticulation and separation in rapid sequences may involve distinct computations: coarticulation possibly involves replacing sub-movements with a motor chunk, while separation possibly indicates independent control of each sub-movement with chunking at a higher-level (Wong and Krakauer, 2019; Zimnik and Churchland, 2021). Thus, there are unresolved questions regarding why sequential movements sometimes coarticulate, and how the representation of future goals in the sensorimotor system influences the way sequences are executed.

Figure 1

Download asset Open asset

Strikingly, most studies have emphasized potential differences in the state of the sensorimotor system during preparatory activity (Zimnik and Churchland, 2021; Ben-Shaul et al., 2004; Hatsopoulos et al., 2003; Lu and Ashe, 2005). As a consequence, the role of feedback control in this debate has been unexplored, although it is known that human motor control involves continuous processing of sensory feedback (Crevecoeur and Kurtzer, 2018; Parrell and Houde, 2019). The theory of optimal feedback control (OFC) has been particularly useful in predicting the influence of numerous task parameters on the controller (Diedrichsen et al., 2010; Cluff and Scott, 2013; Cluff and Scott, 2015; Kalidindi and Crevecoeur, 2023; Todorov and Jordan, 2002; Scott, 2004; Kurtzer et al., 2008; De Comite et al., 2022), thus reproducing goal-directed motor commands during both unperturbed movements and feedback responses to disturbances (Kalidindi and Crevecoeur, 2023). OFC has been used in numerous studies to interpret flexible feedback responses occurring in the long-latency response period (Kalidindi and Crevecoeur, 2023; Scott, 2016). In this framework, all targets in a sequence can be considered simultaneously by the controller, and the overlap between control gains applied to each target determines when they influence the sequence (Figure 1B). The control gains themselves depend on task requirements (Liu and Todorov, 2007; Dimitriou et al., 2013; Cesonis and Franklin, 2021). Thus, OFC predicts two experimentally testable features of sequence production: first, even if the sequence is considered as a whole at the level of controller, control gains can show different amounts of overlap based on task requirements, leading to either coarticulation or separation of the sequence elements; second, in sequences with strong co-articulation, feedback responses to perturbations are expected to depend on future targets due to the overlap of control gains.

Here, we studied these predictions of OFC in a two-reach sequence task. Simulations revealed that the instruction to slow-down at the first target influenced the coarticulation of the two reaches. Importantly, the penalty for the velocity at the first target produced an apparent separation of two reaches, even if the controller was always computed jointly for both reaches. We conducted two experiments with human participants and observed that their behavior was fully consistent with the simulations. We observed both flexible coarticulation dependent on task instruction about the intermediate target, as well as a modulation of long-latency feedback responses to the mechanical loads applied during the first reach dependent on the location of the second target when coarticulation was present. These results demonstrate that both coarticulation and separation of sub-movements may arise as the consequence of efficient feedback control. Recordings of muscle activity suggest that this operation is carried out by the sensorimotor network that mediates fast state-feedback control.

The results section is organized as follows: we first introduce the computational model and describe the results of the simulations showing that separate or coarticulated sequence production can result from flexible feedback control without implying a different structure in the controller (Figure 1A–B). Second, we present the two behavioral experiments in which participants were instructed to perform a sequence of two reaching movements with an intermediate goal (Figure 1C–F). Our key manipulation guided by the model simulations was to give them different instructions regarding the pause at the first target, which produced the modulation of sequence production predicted in theory.

Simulations of sequence execution

The model considers a task-dependent state-feedback controller derived from the framework of stochastic optimal control (Kalidindi and Crevecoeur, 2023). The controller was optimized to produce a translation of a point mass on a horizontal plane. A given estimate of a state vector ${\displaystyle \hat{x}}$ (position, velocity, target location and external load) can influence the motor output $u$ through multiplicative functions called ‘feedback control gains,’ represented by $k\left(t\right)$. Typically, the motor output at any time $\left(t\right)$ is a result of a state feedback control law:

$u\left(t\right)=k(t)\hat{x}(t).$

Note that the control law is dependent on the state estimate (${\displaystyle \hat{x}}$), which is computed by integrating delayed and noisy sensory feedback ($x$) with internal prediction of the state. Importantly, feedback gains are tuned to task demands (Liu and Todorov, 2007; Nashed et al., 2012), and vary within a movement (Dimitriou et al., 2013). This task dependency gives rise to instances where all state variables (${\displaystyle \hat{x}}$) are simultaneously available to the feedback controller, but some of them may not influence the motor output if deviations of the state vector in the corresponding dimension do not impact the performance.

We applied this control formalism to a task composed of a two-target sequence of reaching movements, such that two different sequences shared the first target, after which they diverged towards the second target (T1 and T2, Figure 1C). In the model, the estimated positions of the targets (${\displaystyle \widehat{T1}}$, ${\displaystyle \widehat{T2}}$) are in the state estimate ${\displaystyle \hat{x}}$, hence the control law can be decomposed and expressed in the following equation (also see Figure 2A):

$u\left(t\right)=k_1(t)\widehat{T1}+k_2(t)\widehat{T2}+k_x(t)\hat{x}(t)$

Figure 2 with 4 supplements see all

Download asset Open asset

where ($k_1$, ${\displaystyle k_2}$) are the feedback gains that determine how much and when each target (${\displaystyle \widehat{T1}}$, ${\displaystyle \widehat{T2}}$) influences the motor output (Figure 2A). This formulation highlights that, even when both targets are simultaneously available to the feedback controller, the overlap between control gains ($k_1$ and $k_2$) will determine if and when the second target influences the first reach (Figure 2B). In the simulations, the control gains and the overlap between the two components of the sequence were not hardcoded but were found as an optimal control solution that improves the efficiency of the sequence as a whole (i.e. as a solution to holistic sequence planning in the sense that all targets are specified to the controller at once). For holistic sequence planning, we formulated the efficiency in terms of a cost function on the terminal movement errors and speed of reaching at both targets (T1 and T2), and total motor output across movement duration (see a model with terminal costs in Methods).

The model simulations revealed that, even though the sequence was planned as a whole, the necessity to slow-down at the first target, which was induced by an intermediate cost on speed at the corresponding time, influenced when the first reach became sensitive to the second target (Figure 2C and D), i.e., when the two reaches in the sequence coarticulate. When there was no constraint on speed at the first target, the simulated hand reached the first target with a high velocity (Figure 2C, left panel), and subsequently continued towards the second target at a high initial speed of >0.2 m/s. In this rapid transfer scenario, the control gains showed a cross-over effect where the control gain associated with the second target (k2) was non-zero, albeit small, during the first reach, showing an early overlap with the feedback gain (k1) (Figure 2C, middle panel). As a result, the location of the second target influenced the lateral hand deviation of the first reach (Figure 2C, right panel). The first reach was thus different depending on the location of the second target. Notably, the first reach was similarly influenced by the second target even in a different model where temporal buildup costs were considered instead of terminal costs (see temporal buildup cost model in Methods), to penalize the hand deviation from the targets across the duration of the sequence (Figure 2—figure supplement 1A).

Because the control gains also determine the response to external perturbations, the trajectories following external loads applied to the mass were also dependent on the second target (Figure 2—figure supplement 2A). Interestingly, the lateral deviation along the perturbed direction was smaller when the second target was in the direction of perturbation. Although this may appear counterintuitive, the model shows that it is an efficient solution, because a higher acceleration linked to reversal at T1, when the perturbation is in the direction of T2, incurs a higher motor cost. Hence, a smaller hand deviation is necessary to avoid high acceleration at T1. This result qualitatively persisted across broad variations in parameters related to the costs, delays, and point-mass dynamics (not shown). A similar result was observed when we used temporal buildup of kinematic error in the cost function (Figure 2—figure supplement 2B), instead of using the simple terminal costs, matching human data closely (see the experimental results).

On the contrary, the requirement to slow down at the first target, captured by an increased cost on speed, reduced the velocity at the end of the first reach and altered the shape of the trajectories (Figure 2D, left panel) compared to the rapid transfer scenario (Figure 2C, left panel). The control gain corresponding to the second target became non-zero at a later time, at the end of the first reach (Figure 2D, middle panel). Due to the later increase in this gain value, the influence of the second target on the motor output during the first reach was much reduced, hence generating a similar hand path to reach the first target across the two sequences (Figure 2D, right panel). Similarly, feedback responses to lateral perturbations that were applied early-on in the first reach were not dependent on the location of the second target (Figure 2—figure supplement 3A). The amount of lateral deviation when an unexpected perturbation was applied did not vary as a function of the direction of the second target. Qualitatively similar influence of the need to slow down at the first target on the trajectory of first reach was observed when temporal buildup costs were included instead of terminal errors (Figure 2—figure supplement 1B and Figure 2—figure supplement 3B).

It is worth noting that the OFC model can be generalized to longer sequences (Kashefi et al., 2024) through the incorporation of additional cost terms (in Equation 10 of Methods) and targets, enabling simultaneous planning for more than two targets. Simulations of a sample three-reach sequence (Figure 2—figure supplement 4) revealed that, varying the cost of dwell velocity at intermediate targets ($w_2$ and ${\displaystyle w_3}$ parameters in Methods) caused a variation in control gains. Different amounts of change in control gains can be expected for intermediate versus late targets (Figure 2—figure supplement 4A). Notably, even when we used the same dwell velocity cost ($w_2$ = $w_3$ = 0), the observed velocity profiles were different between the two sequences towards different final targets (T3 up and T3 down) (Figure 2—figure supplement 4B). We tested a condition in which both sequence reaches were forced to have similar dwell velocity profiles by increasing the dwell velocity costs in the sequence towards one of the targets (T3 down), while leaving this parameter unchanged for the other target (T3 up). In this scenario, T3 up sequence had the parameters ($w_2$, $w_3$) = (0, 0), while T3 down sequence had the parameters (0.8, 0.8). In this case, the curvature of the first reach was different, and predominantly occurred due to differences in $K_2$ between the two sequence reaches (Figure 2—figure supplement 4C). These simulations highlight that, planning for a longer horizon sequence can indirectly influence the curvature of early reaches, due to the interaction between intermediate dwell constraints, spatial arrangement of targets, and sequence horizon in a task-dependent manner.

Overall, the model predicted that even if a feedback control policy was computed by optimizing the whole sequence over a long time-horizon, the requirements associated with intermediate goals determine how early in the sequence the second (future) target can influence the feedback controller (compare Figure 2C and Figure 2D; Figure 2—figure supplement 1A and Figure 2—figure supplement 1B; Figure 2—figure supplement 2 and Figure 2—figure supplement 3).

Correspondence between human behavior and model predictions

The experiments were designed to verify whether task requirements at the intermediate goals influenced hand trajectories and feedback control in a way that was expected from the model simulations. The first experiments tested the scenario in which there was no penalty on velocity during the stop-over at the first target, referred to as the GO task. We instructed 15 human participants to perform a two-target sequence of reaching movements (Figure 1C–F). Both targets in a given sequence were displayed from the beginning of the trial, allowing for the preparation of both reaches. After receiving a GO cue (500–1000 ms after target display), participants were required to reach the first target within a prescribed time (500–650 ms), and then to proceed to the second target. There was no instruction to stop at the first target. To encourage fast transfer between the two reaches, the participants were allowed to remain in the first target for a maximum of 150 ms after a successful reach, before initiating the second reach, otherwise the second target disappeared. To determine if the feedback controller was tuned to the location of the second target, mechanical loads (spanning 200ms width) in randomized direction (rightward or leftward) were applied randomly in 20% of the trials soon after the hand left the starting location (4.5 cm) in forward direction from the starting location (this occurred typically at ~200 ms after the movement onset).

The model prediction about the influence of the second target during the first reach was clearly borne out of the experimental results. Figure 3A illustrates the average hand speed across baseline trials in the rapid transfer scenario. It is clear that the hand did not come to rest at the end of the first reach, but continued towards the second target at a relatively high transfer speed of ~0.2 m/s. The hand speed at the first target was slightly different between the 2 s target conditions (Figure 3A, inset; paired t-test: t(14) = –3.8, p=0.002). In line with model predictions, in the rapid transfer scenario, the location of the second target influenced the hand path during the first reach (Figure 3B). Statistically, across trials and participants, lateral hand deviation during the first reach was significantly different between the two sequences (Figure 3B, inset; paired t-test: t(14) = –5.9, p<10^–4).

Figure 3 with 1 supplement see all

Download asset Open asset

Note that the experiments involve a block design where a sequence direction was maintained for 10 trials (with two randomly interleaved rightward and leftward perturbation trials), while the blocks themselves were randomly interleaved (Figure 1E). To test whether the influence of the second target on hand deviation was due to trials occurring late within a block, we computed the hand path in the first and last trials and averaged across all blocks for each participant. Group data in Figure 3—figure supplement 1 illustrate that the hand deviation was significantly different between the leftward and rightward sequence conditions from the very first trial within a block (paired t-test: t(14) = 5.7, p<10^–3). Similarly, the hand deviation was significantly different between leftward and rightward sequences in the last trial within a block (paired t-test: t(14) = 5.9, p<10^–3). Additionally, a temporal effect was observed in the leftward sequence condition, where the last trial incurred a higher hand deviation compared to the first trial within blocks (paired t-test: t(14) = 3.9, p=0.003). Note that the above results include correcting for multiple comparisons using the Holm-Bonferroni method. Overall, the block design of the experiment may lead to small differences between the first and last trials within a block, but the effect of the second target on the lateral hand deviation was present from the very first trial within a block, suggesting that qualitatively similar results can be expected in experiments without a block design.

During perturbation trials, hand trajectories during the first reach were influenced by the location of the second target. In rightward perturbation trials, the dwell velocity and hand deviation (Figure 3C and D) were significantly different across the 2 s target conditions (paired t-test: t(14) = –7.1, p<10^–5 for dwell velocity; t(14) = –7.9, p<10^–5 for hand deviation). Similar differences across the two sequences were observed in the leftward perturbation trials (Figure 3E and F) (paired t-test: t(14) = 6.9, p<10^–5 for dwell velocity; t(14) = –8.2, p<10^–6 for hand deviation). The influence of the second target on the lateral hand deviation was qualitatively similar to that observed in model simulations, and counterintuitive to what we might expect without the help of the model simulations. As observed in the model simulations (see also Figure 2—figure supplement 2), lateral hand deviation was smaller when the perturbation was in the direction of the second target (T2) and vice-versa. This was consistent for both rightward and leftward perturbation conditions. Both the model and humans expressed this strategy that can be seen as an emergent feature of efficient feedback control during the production of movement sequences. Additionally, even though behavior was reproduced in simulations, changing the cost of control effort and/or accuracy of intermediate reaches could modulate the sequence-dependent changes in curvature.

We identified a pair of muscles in the upper-arm shoulder joint (Pectoralis Major, PEC, and Posterior Deltoid, PD) that were strongly recruited by the lateral perturbation (see earlier studies with a similar experimental setup De Comite et al., 2022; Crevecoeur et al., 2019). These muscles also steer the hand toward the first and second targets in unperturbed conditions. PEC muscle displayed agonist burst coinciding with the acceleration of the hand towards the first target, while the PD muscle provided the antagonist burst to decelerate the hand near the first target. Notably, both muscles displayed a clear difference in baseline EMG activity during the first reach, depending on whether the following reach involved right or left second targets (Figure 3G). Following the perturbation (left or right), the stretch responses were visibly larger in both PEC and PD muscles compared to the baseline response (Figure 3H, I). The perturbation response in the stretched muscle – PEC for the rightward load (Figure 3H, left panel) and PD muscle for the leftward load (Figure 3I, right panel) – was visibly larger if the second reach involved movement reversal compared to the first reach where the perturbation was in the direction of the second target. Interestingly, even the unloading response (below baseline) in the unstretched (shortened) muscle – PD for the rightward load (Figure 3H, right panel) and PEC for the leftward load (Figure 3I, left panel) – was larger if the second reach involved direction reversal. Overall, the perturbation responses – both stretch and shortening muscle responses – indicated vigorous feedback corrections against the load when the second target induced movement reversal at the first target, explaining the smaller lateral deviation of the hand when the perturbation was in the direction of the second target.

Sequence-dependent long-latency feedback responses

We assessed when the muscle stretch and shortening responses showed dependency on the second target. If the two targets were simultaneously used to form an efficient feedback controller for the whole sequence, one would expect that the sensitivity of the response to the second target to become visible in the long-latency epoch, considering responses in this epoch as a proxy of flexible, state-feedback controller in the primate nervous system (Kalidindi and Crevecoeur, 2023; Scott, 2016). Overall, the EMG responses indicated a significant influence of the second target on the long latency responses in both stretched and shortened muscles (Figure 4). When perturbed in the rightward direction, the PEC muscle (agonist) response increased in both sequence conditions, which counters the perturbation by increasing the leftward pull. Given the lower hand deviation in the rightward sequence (Figure 3D), we asked if the reduced deviation was due to a larger leftward pull from the PEC muscle. Indeed the PEC stretch response was significantly larger in the rightward sequence than the leftward sequence in the long-latency epoch, in both R2 (one tail paired t-test: t(14) = 3.5, P<0.002) and R3 (one tail paired t-test: t(14) = 2.25, P=0.02). On the other hand, the PD muscle shortening response was significantly larger (below baseline) in the rightward sequence than the leftward sequence in both R2 (one-tailed paired t-test: t(14) = –3.2, P=0.003) and R3 epochs (one-tailed paired t-test: t(14) = –3.5, P=0.002).

Figure 4

Download asset Open asset

When the perturbation load was to the left, the PEC shortening response (below baseline) was significantly larger in the leftward sequence compared to the rightward sequence in both R2 (one tail t-test: t(14) = 7, P<10^–5) and R3 epochs (t(14) = 8.1, P<10^–6). Such a large unloading response in the PEC muscle could facilitate a larger correction in the leftward sequence by reducing the leftward pull in the direction of the applied load. While the shortening response in PEC was larger, the stretch response in PD muscle was not significantly larger in the leftward sequence throughout the long-latency time period (one-tail paired t-test; R2 epoch: t(14) = 2.8, P=0.99; R3 epoch: t(14) = –0.13, P=0.44). On the contrary, the PD stretch response was significantly smaller in the leftward sequence in the R2 epoch (one-tailed t-test: t(14) = 2.8, P=0.006). All muscles exhibited differences in the voluntary time window (120–180ms) during the first reach depending on the location of the second target (two-tailed paired t-test: t (14) absolute value between (3.5–7.4), and p-value between (1e^–6 – 0.012) across muscles and perturbation types).

Next, we verified that the differences reported above did not follow from mere differences in baseline muscle activities at perturbation onset. Muscle responses exhibit a property called ‘automatic gain scaling’ where the stretch response increases proportionally to the pre-perturbation muscle activity (Pruszynski et al., 2009). If gain-scaling was significant, then we expected to observe a difference in the R1 period response between the two sequence conditions. We did not observe any sequence-related modulation of the stretch responses in the R1 epoch, suggesting that any potential differences at baseline did not have a strong impact. It is also important to realize that the sequence direction that involved a larger stretch response in the corresponding muscle was always associated with lower levels of activity in the baseline condition, thus potential biases induced by baseline differences were in the direction opposite to the effect reported above.

Overall, the results indicate, the when required to perform movement sequences rapidly, the sensorimotor circuits that mediate long latency feedback responses showed modulation that was dependent on the secondary target. The specificity of the controller to the second target was evident when perturbations were applied early-on during the first reach. These observations agree with the predictions of the feedback control model that, with low or absent penalty on the velocity at the first target, the second target influences the hand path and feedback responses to perturbations prior to the first target.

Separation of Sequence Elements

Another clear prediction of the model is that when there is a cost on velocity at the first target, the first reach should be less dependent on the second target. To test this prediction experimentally, we performed a STOP task where 14 subjects were asked to perform the same two-target reaching movements with a different speed constraint (Figure 1F). They were required to bring the hand to a near-complete halt (speed <0.5 cm/s) before starting the second reach.

Figure 5 illustrates that the hand speed at the first target dropped to near zero levels in all unperturbed, leftward, and rightward perturb conditions. Clearly, there was no influence of the second target on the velocity of reaching the first target (two-tailed paired t-test: t(13) = –1.9, P=0.08 in the baseline condition; t(13) = –0.01, P=0.99 for rightward perturbations; t(13) = 2.1, P=0.055 for leftward perturbations). The hand deviation during the first reach was similar between the two sequences (two-tailed paired t-test: t(13) = –1.8, P=0.08 in the baseline condition; t(13) = –1.2, P=0.26 for rightward perturbations; t(13) = –1.2, P=0.25 for leftward perturbations). Moreover, the unperturbed EMG responses of the PEC and PD muscles were similar throughout the first reach, and diverged only after the end of the first reach, while moving toward the second targets (T2). In rightward and leftward perturbation conditions, the perturbation responses (ΔEMG) of the PEC and PD muscles (Figure 5D and E, and Figure 6) were not statistically different in pre-perturbation, R1, R2, or R3 epochs. Only the PEC perturbation response, to a leftward perturbation, showed a weak effect of the second target location (R2 epoch: t(13) = 2.1, p=0.055; R3 epoch: t(13) = 2.44, p=0.03). To conclude, these results suggested that the second target did not influence the first reach when the hand was required to slow-down at the first target.

Figure 5

Download asset Open asset

Figure 6

Download asset Open asset

A straightforward interpretation could be that the stopping at the first target invoked a completely different strategy in which the control of the two reaches was performed independently (Figure 1A), effectively separating the two movements, whereas executing them rapidly could produce the merging of the two sub-movements into a coarticulated sequence. While this is conceptually valid, it is not necessary and the model provides a more nuanced view: both apparent separation or coarticulation of the two motor patterns can be explained within the same framework of flexible feedback control. These different modes of sequence execution still require proper specification of the task constraints in the model, such as a number of intermediate steps, dwell-time, or velocity limit. Such specifications must be considered as input to the controller.

The results presented here are consistent with recent behavioral studies showing that the information about future targets in a sequential task is integrated at the stage of motor planning (Diamond et al., 2017; Gottwald et al., 2018; Kashefi et al., 2024; Ariani et al., 2021; Engel et al., 1997). Our key contribution is twofold: we showed from computational and behavioral perspectives that the coarticulation and separation of sequence components depended on task instructions, and likely reflected efficient solutions of the same continuous control problem with distinct penalties at the intermediate goals; second, the results highlighted that long-latency feedback control was modulated by future targets in a coarticulated sequence, consistent with theories suggesting that movement planning is synonymous with derivation of a control policy (Crevecoeur et al., 2014; Wong et al., 2015).

A key question the model and experiments addressed was whether the integration of multiple goals required processing in upstream voluntary circuits, indicating timed, voluntary recall of future goals (Zimnik and Churchland, 2021; Schimel et al., 2023; Logiaco et al., 2021). Instead, our results support the idea of flexible and simultaneous processing of sequence elements, where a low-level controller plays a crucial role in combining these elements. Our model based on optimal feedback control showed that even if two reaches are planned as a whole, the second target does not necessarily influence the controller early. The constraints placed on the swiftness of transition around the first target modulated when the second target started to influence the sequence. This was clearly apparent in the hand deviation during the first reach, and further revealed by perturbing participants’ hands with external loads, evoking sequence-dependent long-latency responses in Experiment 1. Consistent with this model (Figure 1B), recently an artificial neural network model with simultaneous processing of targets reproduced the population activity of the primary motor cortex (see Wang et al., 2024), a region that contributes significantly to long latency feedback responses (Scott, 2016; Pruszynski et al., 2011). Together, these studies point to the importance of considering feedback gains, and suggest a special role of transcortical feedback through M1 in the production of movement sequences. Notably, in the framework of optimal feedback control, an intermediate goal is equivalent to a via-point that constrains the execution of the sequence (similar to Ramkumar et al., 2016). It is thus possible that coarticulation in motor systems be processed similarly as other kinds of movement constraints, such as via-points, avoiding obstacles, or changes in control policies.

Although OFC has been predominantly used as a behavioral-level framework agnostic to neural activity patterns, it can shed light on the planning, state estimation, and execution-related computations in the transcortical feedback pathway (Takei et al., 2021). Using OFC, our study proposes a novel and precise definition of the difference to expect in neural activities in order to identify coarticulated versus independent sequence representations from a computational point of view. Because each condition (i.e. overlapping versus non-overlapping controllers as in Figure 2) was associated with different cost-functions and time-varying control gains, it is the process of deriving these control gains, using the internal representation of the task structure, that may differ across coarticulated and separated sequence conditions. To our knowledge, how and where this operation is performed is unknown. A corollary of this definition is that the preparatory activity (Zimnik and Churchland, 2021; Ames et al., 2019) may not discern independently planned or coarticulated sequences because these situations imply different control policies (and cost functions), as opposed to different initial states. Moreover, the nature of the sequence representation is potentially not dissociable from its execution for the same reason.

Many studies highlight a phenomenon called chunking, where a subset of individual elements of a sequence are concatenated and represented as a unified whole called chunks (Ramkumar et al., 2016; Verwey, 2001; Bera et al., 2021). In the pursuit of finding motor chunks, recent neural recordings explored for holistic (or) unified representations in the circuits that produce motor commands, and found no such representations (Wong and Krakauer, 2019; Yokoi and Diedrichsen, 2019; Zimnik and Churchland, 2021; Berlot et al., 2020). Hence it was suggested that chunking resulted from a cognitive or high-level recall of intermediate goals sent to the controller for independent execution (Wong and Krakauer, 2019, see Figure 1A, but, see Huberdeau et al., 2023). On the contrary, our developments also provide a computational definition of motor chunks. From our model, motor chunking (or a unified representation) does not necessarily imply a single (unified) preparatory period signal representing multiple goals. Motor chunking may represent the formation of a composite cost function, hence a composite control policy, that can jointly optimize multiple sub-movements (see Figure 1B). In this view, motor chunking results from an acquired motor ability to form overlapping control policies for contiguous sequence elements (Figure 1B and Figure 2). Importantly this property depends on the instruction at intermediate goals, thus even a motor chunk may falsely appear to be independently controlled (Figure 1A), masking the signature of sequence representation in neural data even if the controller was formed at once by considering multiple goals.

There are differences between the model and human movements that require further study, such as the influence of sequence direction on the dwell velocity, which was not observed in the model simulations (compare velocity plots in Figure 2C and Figure 3A). A limitation is that nonlinear dynamics were ignored in our model. However, the potential impact of non-linear mechanics on limb trajectory could be completely suppressed by participants in Experiment 2, since we observed a straight reach path, thus making the curvature observed in Experiment 1 more likely related to the task constraints rather than the biomechanics which did not differ between experiments. Yet, the quantification of how biomechanical factors influence control may deserve a dedicated work, as it has been shown that biomechanics can also influence coarticulation in a speech production task (Ostry et al., 1996). Despite the limitations, the simple linearized OFC models were crucial in determining the relationship between three properties of the task demands, its impact on coarticulation, and feedback control. Our approach is consistent with how ideal actor/estimator models have been predominantly employed as normative models in formalizing an initial hypothesis, and simulating these to generate testable predictions (De Comite et al., 2022; Liu and Todorov, 2007; Cesonis and Franklin, 2021; Nashed et al., 2012; Crevecoeur et al., 2019; Takei et al., 2021; Diedrichsen, 2007; Crevecoeur et al., 2020b; Izawa et al., 2008; Ikegami et al., 2021; De Comite et al., 2023; Mistry et al., 2013). Such normative models should be distinguished from detailed biomechanical models that can trade-off interpretability and predictability for achieving better quantitative fits. Furthermore, the main feature of the model – to generate a task-dependent motor output as a combination of state feedback – is not limited to linear controllers and linear body dynamics, but is applicable even when the control policy is non-linear (see Kalidindi et al., 2021; Lillicrap and Scott, 2013).

The modeling and empirical results presented here describe the computations underlying a two-reaches sequence. Although two is admittedly the lowest possible length of a non-trivial sequence, it enables comparison with similar tasks that have been used in recent neurophysiological studies (Zimnik and Churchland, 2021; Ben-Shaul et al., 2004; Hatsopoulos et al., 2003; Wang et al., 2024) to understand the neural computations underlying the control of sequences. We expect that longer sequences be formed in a similar way as we showed that there is no theoretical constraint on the number of intermediate goals in the model and the addition of sequence elements is straightforward. Thus, we expect the combination of two elements can be the basis of longer sequence formation constructed on the same principles.

Besides the implication of our computational model of co-articulation, it has been often documented that swift execution of sequences was achieved by habituation, where a set of sequences are practiced for several days to months (Ramkumar et al., 2016; Yokoi and Diedrichsen, 2019; Zimnik and Churchland, 2021; Lu and Ashe, 2005; Berlot et al., 2020; Huberdeau et al., 2023; Ariani and Diedrichsen, 2019; Maceira-Elvira et al., 2022). Such a process potentially involves a slow time-scale linked to the acquisition of a motor skill (Maceira-Elvira et al., 2022; Beukema and Verstynen, 2018), which differs from the results presented here, as the participants were able to execute accurate sequences after a pre-training of around 100 repetitions. Here, the participants received explicit information about the dwell constraints at the intermediate goal, resulting in quick coarticulation (also see Sosnik et al., 2004; Kashefi et al., 2024; Friedman and Korman, 2019; Sosnik et al., 2007; Sporn et al., 2020). This is in line with task-dependent modulation of existing reaching controllers, which can be rapidly switched from one trial to another (De Comite et al., 2022; Cesonis and Franklin, 2022). This suggests that there exist several mechanisms operating at different time-scales and responsible for a combination of sequence elements. One mechanism is a slow habituation process in which the subject learns de novo to build a novel controller for different elements in the sequence (Huberdeau et al., 2023; Yang et al., 2021), whereas the other mechanism is a fast modulation of this consolidated controller by task demands. In principle, the two mechanisms are dissociable with perturbations, which may allow future studies to investigate how they interact and to identify which brain regions mediate these components of skilled motor behavior.

The data from simulations and experiments is available in the public repository 'figshare'.

1. 1. Kalidindi H
   2. Crevecoeur F

   (2024) figshare

   Task dependent coarticulation of motor sequences.

   https://doi.org/10.6084/m9.figshare.24282418

1. 1. Ames KC
   2. Ryu SI
   3. Shenoy KV

   (2019) Simultaneous motor preparation and execution in a last-moment reach correction task

   Nature Communications 10:2718.

   https://doi.org/10.1038/s41467-019-10772-2

   • PubMed
   • Google Scholar
2. 1. Ariani G
   2. Diedrichsen J

   (2019) Sequence learning is driven by improvements in motor planning

   Journal of Neurophysiology 121:2088–2100.

   https://doi.org/10.1152/jn.00041.2019

   • PubMed
   • Google Scholar
3. 1. Ariani G
   2. Kordjazi N
   3. Pruszynski JA
   4. Diedrichsen J

   (2021) The planning horizon for movement sequences

   eNeuro 8:ENEURO.0085-21.2021.

   https://doi.org/10.1523/ENEURO.0085-21.2021

   • PubMed
   • Google Scholar
4. 1. Ben-Shaul Y
   2. Drori R
   3. Asher I
   4. Stark E
   5. Nadasdy Z
   6. Abeles M

   (2004) Neuronal activity in motor cortical areas reflects the sequential context of movement

   Journal of Neurophysiology 91:1748–1762.

   https://doi.org/10.1152/jn.00957.2003

   • PubMed
   • Google Scholar
5. 1. Bera K
   2. Shukla A
   3. Bapi RS

   (2021) Motor chunking in internally guided sequencing

   Brain Sciences 11:292.

   https://doi.org/10.3390/brainsci11030292

   • PubMed
   • Google Scholar
6. 1. Berlot E
   2. Popp NJ
   3. Diedrichsen J

   (2020) A critical re-evaluation of fMRI signatures of motor sequence learning

   eLife 9:e55241.

   https://doi.org/10.7554/eLife.55241

   • PubMed
   • Google Scholar
7. 1. Beukema P
   2. Verstynen T

   (2018) Predicting and binding: interacting algorithms supporting the consolidation of sequential motor skills

   Current Opinion in Behavioral Sciences 20:98–103.

   https://doi.org/10.1016/j.cobeha.2017.11.014

   • Google Scholar
8. 1. Botvinick M
   2. Plaut DC

   (2004) Doing without schema hierarchies: a recurrent connectionist approach to normal and impaired routine sequential Action

   Psychological Review 111:395–429.

   https://doi.org/10.1037/0033-295X.111.2.395

   • Google Scholar
9. Preprint

   1. Cesonis J
   2. Franklin DW

   (2021) Mixed-horizon optimal feedback control as a model of human movement

   arXiv.

   http://arxiv.org/abs/2104.06275

   • Google Scholar
10. 1. Cesonis J
    2. Franklin DW

    (2022) Contextual cues are not unique for motor learning: Task-dependant switching of feedback controllers

    PLOS Computational Biology 18:e1010192.

    https://doi.org/10.1371/journal.pcbi.1010192

    • PubMed
    • Google Scholar
11. 1. Cluff T
    2. Scott SH

    (2013) Rapid feedback responses correlate with reach adaptation and properties of novel upper limb loads

    The Journal of Neuroscience 33:15903–15914.

    https://doi.org/10.1523/JNEUROSCI.0263-13.2013

    • PubMed
    • Google Scholar
12. 1. Cluff T
    2. Scott SH

    (2015) Apparent and actual trajectory control depend on the behavioral context in upper limb motor tasks

    The Journal of Neuroscience 35:12465–12476.

    https://doi.org/10.1523/JNEUROSCI.0902-15.2015

    • PubMed
    • Google Scholar
13. 1. Crevecoeur F
    2. Sepulchre RJ
    3. Thonnard JL
    4. Lefèvre P

    (2011) Improving the state estimation for optimal control of stochastic processes subject to multiplicative noise

    Automatica 47:591–596.

    https://doi.org/10.1016/j.automatica.2011.01.026

    • Google Scholar
14. Book

    1. Crevecoeur F
    2. Cluff T
    3. Scott SH

    (2014) Computational Approaches for Goal-Directed Movement Planning and Execution

    MIT press.

    https://doi.org/10.7551/mitpress/9504.003.0051

    • Google Scholar
15. 1. Crevecoeur F
    2. Kurtzer I

    (2018) Long-latency reflexes for inter-effector coordination reflect a continuous state feedback controller

    Journal of Neurophysiology 120:2466–2483.

    https://doi.org/10.1152/jn.00205.2018

    • PubMed
    • Google Scholar
16. 1. Crevecoeur F
    2. Scott SH
    3. Cluff T

    (2019) Robust control in human reaching movements: a model-free strategy to compensate for unpredictable disturbances

    The Journal of Neuroscience 39:8135–8148.

    https://doi.org/10.1523/JNEUROSCI.0770-19.2019

    • PubMed
    • Google Scholar
17. 1. Crevecoeur F
    2. Mathew J
    3. Bastin M
    4. Lefèvre P

    (2020a) Feedback adaptation to unpredictable force fields in 250 ms

    eNeuro 7:ENEURO.0400-19.2020.

    https://doi.org/10.1523/ENEURO.0400-19.2020

    • PubMed
    • Google Scholar
18. 1. Crevecoeur F
    2. Thonnard JL
    3. Lefèvre P

    (2020b) A very fast time scale of human motor adaptation: within movement adjustments of internal representations during reaching

    eNeuro 7:ENEURO.0149-19.2019.

    https://doi.org/10.1523/ENEURO.0149-19.2019

    • PubMed
    • Google Scholar
19. 1. De Comite A
    2. Crevecoeur F
    3. Lefèvre P

    (2022) Continuous tracking of task parameters tunes reaching control online

    eNeuro 9:ENEURO.0055-22.2022.

    https://doi.org/10.1523/ENEURO.0055-22.2022

    • PubMed
    • Google Scholar
20. 1. De Comite A
    2. Lefèvre P
    3. Crevecoeur F

    (2023) Continuous evaluation of cost-to-go for flexible reaching control and online decisions

    PLOS Computational Biology 19:e1011493.

    https://doi.org/10.1371/journal.pcbi.1011493

    • PubMed
    • Google Scholar
21. 1. Diamond JS
    2. Wolpert DM
    3. Flanagan JR

    (2017) Rapid target foraging with reach or gaze: The hand looks further ahead than the eye

    PLOS Computational Biology 13:e1005504.

    https://doi.org/10.1371/journal.pcbi.1005504

    • PubMed
    • Google Scholar
22. 1. Diedrichsen J

    (2007) Optimal task-dependent changes of bimanual feedback control and adaptation

    Current Biology 17:1675–1679.

    https://doi.org/10.1016/j.cub.2007.08.051

    • PubMed
    • Google Scholar
23. 1. Diedrichsen J
    2. Shadmehr R
    3. Ivry RB

    (2010) The coordination of movement: optimal feedback control and beyond

    Trends in Cognitive Sciences 14:31–39.

    https://doi.org/10.1016/j.tics.2009.11.004

    • Google Scholar
24. 1. Diedrichsen J
    2. Kornysheva K

    (2015) Motor skill learning between selection and execution

    Trends in Cognitive Sciences 19:227–233.

    https://doi.org/10.1016/j.tics.2015.02.003

    • Google Scholar
25. 1. Dimitriou M
    2. Wolpert DM
    3. Franklin DW

    (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
26. 1. Engel KC
    2. Flanders M
    3. Soechting JF

    (1997) Anticipatory and sequential motor control in piano playing

    Experimental Brain Research 113:189–199.

    https://doi.org/10.1007/BF02450317

    • PubMed
    • Google Scholar
27. 1. Fowler CA
    2. Saltzman E

    (1993) Coordination and coarticulation in speech production

    Language and Speech 36 (Pt 2-3):171–195.

    https://doi.org/10.1177/002383099303600304

    • PubMed
    • Google Scholar
28. 1. Friedman J
    2. Korman M

    (2019) Observation of an expert model induces a skilled movement coordination pattern in a single session of intermittent practice

    Scientific Reports 9:4609.

    https://doi.org/10.1038/s41598-019-40924-9

    • PubMed
    • Google Scholar
29. Preprint

    1. Gottwald JM
    2. De Bortoli Vizioli A
    3. Lindskog M
    4. Nyström P
    5. Ekberg TL
    6. von Hofsten C
    7. Gredebäck G

    (2018) Infants Prospectively Control Reaching Based on the Difficulty of Future Actions – To What Extent Can Infants’ Multiple Step Actions Be Explained by Fitts’ Law?

    PsyArXiv.

    https://doi.org/10.31234/osf.io/7wek4

    • Google Scholar
30. 1. Hatsopoulos NG
    2. Paninski L
    3. Donoghue JP

    (2003) Sequential movement representations based on correlated neuronal activity

    Experimental Brain Research 149:478–486.

    https://doi.org/10.1007/s00221-003-1385-9

    • PubMed
    • Google Scholar
31. 1. Hikosaka O
    2. Nakahara H
    3. Rand MK
    4. Sakai K
    5. Lu X
    6. Nakamura K
    7. Miyachi S
    8. Doya K

    (1999) Parallel neural networks for learning sequential procedures

    Trends in Neurosciences 22:464–471.

    https://doi.org/10.1016/s0166-2236(99)01439-3

    • PubMed
    • Google Scholar
32. Preprint

    1. Huberdeau DM
    2. Haith AM
    3. Krakauer JW

    (2023) Continuous motor skills as flexible control policies: a video game study

    bioRxiv.

    https://doi.org/10.1101/2023.09.21.558913v1

    • Google Scholar
33. 1. Ikegami T
    2. Ganesh G
    3. Gibo TL
    4. Yoshioka T
    5. Osu R
    6. Kawato M

    (2021) Hierarchical motor adaptations negotiate failures during force field learning

    PLOS Computational Biology 17:e1008481.

    https://doi.org/10.1371/journal.pcbi.1008481

    • PubMed
    • Google Scholar
34. 1. Izawa J
    2. Rane T
    3. Donchin O
    4. Shadmehr R

    (2008) Motor adaptation as a process of reoptimization

    The Journal of Neuroscience 28:2883–2891.

    https://doi.org/10.1523/JNEUROSCI.5359-07.2008

    • PubMed
    • Google Scholar
35. 1. Kadmon Harpaz N
    2. Hardcastle K
    3. Ölveczky BP

    (2022) Learning-induced changes in the neural circuits underlying motor sequence execution

    Current Opinion in Neurobiology 76:102624.

    https://doi.org/10.1016/j.conb.2022.102624

    • Google Scholar
36. 1. Kalidindi HT
    2. Cross KP
    3. Lillicrap TP
    4. Omrani M
    5. Falotico E
    6. Sabes PN
    7. Scott SH

    (2021) Rotational dynamics in motor cortex are consistent with a feedback controller

    eLife 10:e67256.

    https://doi.org/10.7554/eLife.67256

    • PubMed
    • Google Scholar
37. 1. Kalidindi HT
    2. Crevecoeur F

    (2023) Human reaching control in dynamic environments

    Current Opinion in Neurobiology 83:102810.

    https://doi.org/10.1016/j.conb.2023.102810

    • PubMed
    • Google Scholar
38. 1. Kashefi M
    2. Reschechtko S
    3. Ariani G
    4. Shahbazi M
    5. Tan A
    6. Diedrichsen J
    7. Pruszynski JA

    (2024) Future movement plans interact in sequential arm movements

    eLife 13:RP94485.

    https://doi.org/10.7554/eLife.94485

    • PubMed
    • Google Scholar
39. 1. Kurtzer IL
    2. Pruszynski JA
    3. Scott SH

    (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
40. 1. Lillicrap TP
    2. Scott SH

    (2013) Preference distributions of primary motor cortex neurons reflect control solutions optimized for limb biomechanics

    Neuron 77:168–179.

    https://doi.org/10.1016/j.neuron.2012.10.041

    • PubMed
    • Google Scholar
41. 1. Liu D
    2. Todorov E

    (2007) Evidence for the flexible sensorimotor strategies predicted by optimal feedback control

    The Journal of Neuroscience 27:9354–9368.

    https://doi.org/10.1523/JNEUROSCI.1110-06.2007

    • PubMed
    • Google Scholar
42. 1. Logiaco L
    2. Abbott LF
    3. Escola S

    (2021) Thalamic control of cortical dynamics in a model of flexible motor sequencing

    Cell Reports 35:109090.

    https://doi.org/10.1016/j.celrep.2021.109090

    • PubMed
    • Google Scholar
43. 1. Lu X
    2. Ashe J

    (2005) Anticipatory activity in primary motor cortex codes memorized movement sequences

    Neuron 45:967–973.

    https://doi.org/10.1016/j.neuron.2005.01.036

    • PubMed
    • Google Scholar
44. 1. Maceira-Elvira P
    2. Timmermann JE
    3. Popa T
    4. Schmid A-C
    5. Krakauer JW
    6. Morishita T
    7. Wessel MJ
    8. Hummel FC

    (2022) Dissecting motor skill acquisition: Spatial coordinates take precedence

    Science Advances 8:eabo3505.

    https://doi.org/10.1126/sciadv.abo3505

    • PubMed
    • Google Scholar
45. 1. Mathew J
    2. Lefevre P
    3. Crevecoeur F

    (2020) Rapid changes in movement representations during human reaching could be preserved in memory for at least 850 ms

    eNeuro 7:ENEURO.0266-20.2020.

    https://doi.org/10.1523/ENEURO.0266-20.2020

    • PubMed
    • Google Scholar
46. 1. Miller GA

    (1956)

    The magical number seven plus or minus two: some limits on our capacity for processing information

    Psychological Review 63:81–97.

    • PubMed
    • Google Scholar
47. 1. Mistry M
    2. Theodorou E
    3. Schaal S
    4. Kawato M

    (2013)

    Optimal control of reaching includes kinematic constraints

    Journal of Neurophysiology 110:1–11.

    • Google Scholar
48. 1. Mizes KGC
    2. Lindsey J
    3. Escola GS
    4. Ölveczky BP

    (2023) Dissociating the contributions of sensorimotor striatum to automatic and visually guided motor sequences

    Nature Neuroscience 26:1791–1804.

    https://doi.org/10.1038/s41593-023-01431-3

    • PubMed
    • Google Scholar
49. 1. Nashed JY
    2. Crevecoeur F
    3. Scott SH

    (2012) Influence of the behavioral goal and environmental obstacles on rapid feedback responses

    Journal of Neurophysiology 108:999–1009.

    https://doi.org/10.1152/jn.01089.2011

    • PubMed
    • Google Scholar
50. 1. Ostry DJ
    2. Gribble PL
    3. Gracco VL

    (1996) Coarticulation of jaw movements in speech production: is context sensitivity in speech kinematics centrally planned?

    The Journal of Neuroscience 16:1570–1579.

    https://doi.org/10.1523/JNEUROSCI.16-04-01570.1996

    • PubMed
    • Google Scholar
51. 1. Parrell B
    2. Houde J

    (2019) Modeling the role of sensory feedback in speech motor control and learning

    Journal of Speech, Language, and Hearing Research 62:2963–2985.

    https://doi.org/10.1044/2019_JSLHR-S-CSMC7-18-0127

    • PubMed
    • Google Scholar
52. 1. Pruszynski JA
    2. Kurtzer I
    3. Lillicrap TP
    4. Scott SH

    (2009) Temporal evolution of “automatic gain-scaling.”

    Journal of Neurophysiology 102:992–1003.

    https://doi.org/10.1152/jn.00085.2009

    • PubMed
    • Google Scholar
53. 1. Pruszynski JA
    2. Kurtzer I
    3. Nashed JY
    4. Omrani M
    5. Brouwer B
    6. Scott SH

    (2011) Primary motor cortex underlies multi-joint integration for fast feedback control

    Nature 478:387–390.

    https://doi.org/10.1038/nature10436

    • PubMed
    • Google Scholar
54. 1. Ramkumar P
    2. Acuna DE
    3. Berniker M
    4. Grafton ST
    5. Turner RS
    6. Kording KP

    (2016) Chunking as the result of an efficiency computation trade-off

    Nature Communications 7:12176.

    https://doi.org/10.1038/ncomms12176

    • PubMed
    • Google Scholar
55. Preprint

    1. Schimel M
    2. Kao TC
    3. Hennequin G

    (2023) When and why does motor preparation arise in recurrent neural network models of motor control

    bioRxiv.

    https://doi.org/10.1101/2023.04.03.535429v2

    • Google Scholar
56. 1. Scott SH

    (2004) Optimal feedback control and the neural basis of volitional motor control

    Nature Reviews. Neuroscience 5:532–546.

    https://doi.org/10.1038/nrn1427

    • PubMed
    • Google Scholar
57. 1. Scott SH

    (2016) A functional taxonomy of bottom-up sensory feedback processing for motor actions

    Trends in Neurosciences 39:512–526.

    https://doi.org/10.1016/j.tins.2016.06.001

    • PubMed
    • Google Scholar
58. 1. Sheahan HR
    2. Franklin DW
    3. Wolpert DM

    (2016) Motor planning, not execution, separates motor memories

    Neuron 92:773–779.

    https://doi.org/10.1016/j.neuron.2016.10.017

    • PubMed
    • Google Scholar
59. 1. Sosnik R
    2. Hauptmann B
    3. Karni A
    4. Flash T

    (2004) When practice leads to co-articulation: the evolution of geometrically defined movement primitives

    Experimental Brain Research 156:422–438.

    https://doi.org/10.1007/s00221-003-1799-4

    • PubMed
    • Google Scholar
60. 1. Sosnik R
    2. Flash T
    3. Hauptmann B
    4. Karni A

    (2007) The acquisition and implementation of the smoothness maximization motion strategy is dependent on spatial accuracy demands

    Experimental Brain Research 176:311–331.

    https://doi.org/10.1007/s00221-006-0617-1

    • PubMed
    • Google Scholar
61. Preprint

    1. Sporn S
    2. Chen X
    3. Galea JM

    (2020) Reward-based invigoration of sequential reaching

    bioRxiv.

    https://doi.org/10.1101/2020.06.15.152876v1

    • Google Scholar
62. 1. Takei T
    2. Lomber SG
    3. Cook DJ
    4. Scott SH

    (2021) Transient deactivation of dorsal premotor cortex or parietal area 5 impairs feedback control of the limb in macaques

    Current Biology 31:1476–1487.

    https://doi.org/10.1016/j.cub.2021.01.049

    • PubMed
    • Google Scholar
63. 1. Todorov E
    2. Jordan MI

    (2002) Optimal feedback control as a theory of motor coordination

    Nature Neuroscience 5:1226–1235.

    https://doi.org/10.1038/nn963

    • PubMed
    • Google Scholar
64. 1. Verwey WB

    (2001) Concatenating familiar movement sequences: the versatile cognitive processor

    Acta Psychologica 106:69–95.

    https://doi.org/10.1016/s0001-6918(00)00027-5

    • PubMed
    • Google Scholar
65. 1. Verwey WB
    2. Abrahamse EL

    (2012) Distinct modes of executing movement sequences: reacting, associating, and chunking

    Acta Psychologica 140:274–282.

    https://doi.org/10.1016/j.actpsy.2012.05.007

    • PubMed
    • Google Scholar
66. 1. Wang T
    2. Chen Y
    3. Zhang Y
    4. Cui H

    (2024) Multiplicative joint coding in preparatory activity for reaching sequence in macaque motor cortex

    Nature Communications 15:3153.

    https://doi.org/10.1038/s41467-024-47511-1

    • PubMed
    • Google Scholar
67. 1. Wohlgemuth MJ
    2. Sober SJ
    3. Brainard MS

    (2010) Linked control of syllable sequence and phonology in birdsong

    The Journal of Neuroscience 30:12936–12949.

    https://doi.org/10.1523/JNEUROSCI.2690-10.2010

    • PubMed
    • Google Scholar
68. 1. Wong AL
    2. Haith AM
    3. Krakauer JW

    (2015) Motor planning

    The Neuroscientist 21:385–398.

    https://doi.org/10.1177/1073858414541484

    • PubMed
    • Google Scholar
69. 1. Wong AL
    2. Krakauer JW

    (2019) Why are sequence representations in primary motor cortex so elusive?

    Neuron 103:956–958.

    https://doi.org/10.1016/j.neuron.2019.09.011

    • PubMed
    • Google Scholar
70. 1. Yang CS
    2. Cowan NJ
    3. Haith AM

    (2021) De novo learning versus adaptation of continuous control in a manual tracking task

    eLife 10:e62578.

    https://doi.org/10.7554/eLife.62578

    • PubMed
    • Google Scholar
71. 1. Yokoi A
    2. Diedrichsen J

    (2019) Neural organization of hierarchical motor sequence representations in the human neocortex

    Neuron 103:1178–1190.

    https://doi.org/10.1016/j.neuron.2019.06.017

    • PubMed
    • Google Scholar
72. 1. Zimnik AJ
    2. Churchland MM

    (2021) Independent generation of sequence elements by motor cortex

    Nature Neuroscience 24:412–424.

    https://doi.org/10.1038/s41593-021-00798-5

    • PubMed
    • Google Scholar

Author details

1. Hari Teja Kalidindi

   1. Institute for Information and Communication Technologies, Electronics and Applied Mathematics (ICTEAM), Université Catholique de Louvain, Louvain-la-Neuve, Belgium
   2. Institute of Neuroscience (IoNS), Université Catholique de Louvain, Brussels, Belgium

   Contribution

   Conceptualization, Data curation, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-2634-7953

2. Frederic Crevecoeur

   1. Institute for Information and Communication Technologies, Electronics and Applied Mathematics (ICTEAM), Université Catholique de Louvain, Louvain-la-Neuve, Belgium
   2. Institute of Neuroscience (IoNS), Université Catholique de Louvain, Brussels, Belgium
   3. WEL Research Institute, Wavre, Belgium

   Contribution

   Conceptualization, Resources, Supervision, Funding acquisition, Validation, Investigation, Methodology, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   frederic.crevecoeur@uclouvain.be

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-1147-1153

• 709

  views

• 53

  downloads

• 0

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.