RESEARCH DOSSIER
Sensorimotor Integration
Abstract. Sensorimotor integration coordinates sensory information, estimated body state, predicted consequences, and motor commands to support accurate action and adaptation. This critical narrative research monograph examines the history from reafference and early sensory–motor development experiments to formal internal models and contemporary cerebellar and behavioral research. It compares proprioceptive and visual cue combination, predictive control, optimal feedback, state-space learning, implicit recalibration, explicit strategies, grip-force coordination, and adaptations to mechanical and visual perturbations. Evidence is evaluated through experimental design, measurement, feedback timing, generalization, competing computational explanations, and recent research on aging and circuit-level prediction. Human–artificial sensorimotor coordination is considered as a comparative research context. The synthesis identifies the explanatory importance of distinguishing sensory prediction error from performance error, model fit from mechanism, and short-term adaptation from skill acquisition. It presents a scientific integration of published work rather than a new experiment or meta-analysis.
Keywords: sensorimotor integration; sensorimotor control; proprioception; visual–haptic integration; body-state estimation; forward models; sensory prediction error; optimal feedback control; visuomotor adaptation; motor learning; cerebellum; implicit adaptation; explicit re-aiming; Artificial Era
Sensorimotor Integration as a Psychological Research Object
Sensorimotor integration denotes the ongoing coordination of sensory evidence, estimates of bodily state, motor commands, and the consequences of action. It is investigated when organisms use vision, proprioception, touch, vestibular signals, or other modalities to plan, execute, correct, and learn movements. The scientific object is a coupled process: sensory information changes action, while action alters the information subsequently available. A comprehensive account must explain the organization of feedback and prediction at different timescales, from rapid correction within a reach to recalibration over repeated trials. The original internal-model experiment of Wolpert et al. (1995) placed state estimation at the center of this research tradition, and Shadmehr et al. (2010) later synthesized the role of delayed feedback and predictive learning.
Integration has several separable operational meanings. It can mean statistically combining uncertain measurements of the same variable; transforming information between sensory and motor coordinates; using predicted feedback to distinguish self-generated change from external perturbation; selecting task-relevant sensations for online correction; or updating a controller after consistent error. A study demonstrating one of these processes has not automatically demonstrated all of them. For example, combining seen and felt hand location at movement planning does not by itself establish that the same weighting applies during execution. The controlled dissociations described by Sober & Sabes (2003) show why the computation, goal, and moment of measurement need to be specified together.
The central research questions concern what information is represented, how uncertainty and delay alter its influence, where sensory and motor signals interact, which learning signals recalibrate behavior, and how these processes develop or change in disease, aging, and technological contexts. This dossier treats sensorimotor integration as a scientifically observable family of computations and organism-environment relations. It connects motor neuroscience, psychophysics, computational modeling, and behavioral experiments, while leaving general multisensory causal binding and neural action-selection mechanisms to related studies. The analysis is a critical narrative synthesis of published studies, not an original experiment or a quantitative meta-analysis Wolpert et al. (2011) Franklin & Wolpert (2011).
Scientific Definitions and Levels of Explanation
A sensory signal is an observation produced by a receptor system under particular physiological and environmental conditions. A motor command is an outgoing control signal that contributes to muscle activation or other actuator activity. An estimate of body state is a representation or dynamically inferred quantity concerning position, velocity, orientation, load, or other movement-relevant variables. A predicted sensory consequence is the model-derived observation expected if a particular command acts upon a particular estimated state. These definitions allow experimenters to manipulate feedback independently from the actual movement and to ask which discrepancy drives measured correction. The computational discussion of Miall & Wolpert (1996) provides a classic treatment of such forward models.
Three explanatory levels are indispensable. Behavioral descriptions concern trajectories, forces, reaction times, coordination, perceptual reports, and error correction. Computational descriptions concern tasks, uncertainties, costs, state estimates, sensory predictions, and learning rules. Biological descriptions concern receptors, peripheral pathways, spinal responses, cerebellar processing, cortical networks, and their interactions. A motor trajectory can be predicted by multiple computational architectures, and similar neural activation may accompany distinct task solutions. Consequently the strength of an explanation depends on whether manipulations and measurements discriminate among mechanisms, not merely whether the model is intuitively compatible with a movement Scott (2004) Pruszynski & Scott (2012).
Integration should also be distinguished from motor learning and from general motor competence. A participant may integrate sensory inputs effectively in a familiar task yet have difficulty discovering a novel strategy; another participant may adapt to a perturbation with little improvement in a complex skill. The broad treatment of motor learning by Krakauer et al. (2019) cautions against treating laboratory adaptation and sequence learning as exhaustive measures of skill acquisition. For this dossier, skill learning is included where it illuminates the calibration, coordination, and exploitation of sensory information. Full expertise also requires selecting goals, perceiving environmental opportunities, organizing sequences, and practicing under variable demands.
From Reafference to Experimental Sensorimotor Science
Historical analysis begins with the problem of how an organism preserves a useful perceptual world while its own movements transform sensory surfaces. An eye movement changes retinal stimulation; a reach changes hand position and cutaneous contact; locomotion changes optic flow and vestibular input. A central question in twentieth-century physiology was how sensory changes attributable to one's actions could be distinguished from those attributable to external events. Sperry's experimental study of visual inversion belongs to the corollary-discharge tradition Sperry (1950). Later research used the concepts of efference copy and reafference to formulate the same general predictive relation between commands and consequences. These related terms have distinct historical genealogies and should be interpreted in their source-specific contexts.
Developmental experiments further established that the relation between movement and vision matters for organizing visually guided behavior. In the widely discussed active-versus-passive movement design, Held & Hein (1963) investigated movement-produced stimulation in developing kittens. The study became influential because it placed the contingency between an animal's movement and changing sensory input at the center of visual-motor development. A subsequent experiment distinguished elicited extension from accurate guidance of placing responses Hein & Held (1967). These findings are important historically, yet the animal-rearing conditions, behavioral outcomes, and alternative interpretations must be preserved; a single study does not establish a universal rule that all visual development requires the same form of active movement.
The modern experimental program refined this broad insight through perturbations of feedback, limb dynamics, and target mappings. In robotic force-field experiments, Shadmehr & Mussa-Ivaldi (1994) demonstrated how practice changed reaching trajectories under altered mechanical dynamics. In a later experiment, Wolpert et al. (1995) tested state estimation during movements performed in darkness with externally imposed forces. Together these approaches linked movement adaptation, internal-state estimation, and measurable behavior. The historical transition involved increasingly precise experimental control over when errors arise, whether the discrepancy affects performance or sensation, and how response changes persist once perturbations are withdrawn. It also created new demands for distinguishing descriptive modeling from direct neural measurement.
Body-State Estimation and Proprioceptive Signals
Proprioception supplies information related to body configuration, movement, and force through multiple peripheral sources and central processing. Position sense, movement sense, and force-related judgments are distinct experimental outcomes even when they interact during natural behavior. The comprehensive physiological review by Proske & Gandevia (2012) describes the roles of muscle and other sensory signals in representing limb state. For sensorimotor integration, these signals provide a changing estimate of where the body is and what mechanical conditions are acting on it. Their interpretation is contextual: the same sensory input may be weighted differently during a stationary posture, an active reach, or an unexpected mechanical perturbation.
Visual and proprioceptive estimates may disagree, especially in virtual environments or after altered feedback. A useful experiment displaces a seen hand or cursor while leaving the actual limb stationary or moving through a different path. Researchers can then measure reaching direction, final position, perceived hand location, and subsequent recalibration separately. Sober & Sabes (2003) found that visual and proprioceptive contributions differed across computations involved in planning movement vectors and transforming them into joint-based commands. This finding undermines a simplistic claim that an individual has one global fixed weighting for vision versus proprioception. The relevant weights depend on the quantity being estimated and how the estimate will be used.
Measurement introduces additional complications. Joint-position matching can depend on memory, response mechanics, reference-limb reliability, and the choice between active and passive movement. A perceived position report may also change because of recalibration of sensory estimates rather than modification of the motor controller alone. Baseline uncertainty must therefore be assessed rather than inferred from error size. The practical consequence is that proprioceptive tests should specify their modalities, the manipulation of sensory cues, and the relation between perception and performance. The integration of sensory evidence and motor prediction remains a problem of joint inference, for which the functional review of Franklin & Wolpert (2011) provides a useful computational framework.
Vision, Touch, and Reliability-Dependent Combination
Visual information can supply estimates of external target location, object geometry, and the location of the hand or tool. Touch conveys contact events, surface properties, slip, and force-related consequences that may become available only after movement. Sensory integration is therefore temporally structured: vision frequently informs approach, while tactile feedback becomes critical during manipulation. The experimental analysis of lifting by Johansson & Westling (1984) demonstrated the coordinated regulation of grip force and load force and the relevance of prior sensory experience. A companion manipulation of weight and surface structure Westling & Johansson (1984) highlights how information acquired in one lift can influence control of a subsequent lift.
Reliability-weighted cue combination proposes that uncertain estimates should contribute differently according to their precision. In a visual-haptic height-discrimination task, Ernst & Banks (2002) measured unimodal variances and found that combined judgments resembled predictions from a maximum-likelihood model. This is evidence for near-optimal cue integration under those laboratory conditions, with task-specific uncertainty estimates. It does not imply that all human behavior is statistically optimal or that a perceptual cue always receives weight solely from its isolated variance. Dependencies among cues, prior expectations, causal assumptions, strategic attention, and the need for rapid action can all modify behavior.
The distinction between combining senses and coordinating them with action matters in practical experiments. A person judging object size with vision and touch is solving a perceptual estimation problem; the same person lifting the object must also predict consequences and adjust grip before slippage occurs. The former task is informative about cue weighting, the latter about predictive manipulation and feedback control. Johansson & Westling (1984) and Ernst & Banks (2002) illuminate complementary parts of this process. For generalization, researchers need to test whether cue weights established with stationary judgments predict reaching, grasping, and continuous interaction when information arrives at different times and when the costs of errors are asymmetric.
Forward Models, Inverse Models, and Prediction
A forward model estimates the sensory or dynamical consequences expected to follow a candidate motor command from an estimated state. An inverse model, in a standard computational formulation, maps desired changes or outcomes toward commands capable of producing them. The distinction is explanatory: prediction supports anticipatory adjustment and comparison with feedback, while command selection solves a different transformation. Miall & Wolpert (1996) developed influential arguments for the functions of forward models in motor control. An experimentally observed aftereffect or smooth trajectory is compatible with a family of possible architectures, so demonstrations of a specific forward mechanism require manipulations that alter predicted consequences and distinguish them from alternative sources of correction.
The internal-model experiment of Wolpert et al. (1995) examined estimates of unseen hand position after movements under externally imposed forces. Its temporal patterns supported a model-based account of estimation. Later synthetic work by Shadmehr et al. (2010) connected sensory prediction errors to updating internal models as the body or environment changes. The key distinction concerns a mismatch between predicted and experienced sensory consequences. Such a mismatch can occur even when the target was successfully reached, and conversely a task error can arise when the predicted sensory consequence occurred exactly as expected. Experiments which conflate these discrepancies risk attributing learning to the wrong signal.
Internal models have become productive computational tools because they generate testable predictions about how changes in state estimation, predicted feedback, or learning rate should affect behavior. Their existence in a mathematical model, however, does not establish a uniquely localized neural representation. Different networks can approximate the same input-output transformation, and biological implementation may be distributed. A careful explanatory argument therefore connects model predictions to perturbation experiments, lesion or stimulation evidence, and behavior in contexts beyond the training apparatus. Franklin & Wolpert (2011) compares predictive control with other mechanisms, reminding us that movement can also rely on feedback policies and mechanical stabilization.
Feedback, Delay, and the Architecture of Control
Feedback control uses information about the current or recent state to modify ongoing commands. In sensorimotor systems the relevant observations are delayed by receptor transduction, neural transmission, processing, and mechanics. Directly reacting to old information can produce instability or poor timing; prediction provides one way to estimate what the state may be when a corrective command takes effect. The computational analysis of Shadmehr et al. (2010) frames noisy, delayed feedback as a central control problem. Psychophysical experiments can manipulate the latency or continuity of displayed feedback while holding the mechanical movement similar, thereby separating delayed observation from the opportunity to adjust the limb.
Optimal feedback control models a controller that minimizes expected task costs under constraints, using state estimates to guide corrections selectively. This family of models helps explain why deviations that threaten an intended outcome may elicit stronger correction than deviations that leave success possible. Scott (2004) reviews the neural and biomechanical rationale, and Pruszynski & Scott (2012) links task-sensitive long-latency stretch responses to feedback-control principles. These studies invite an integrated account of voluntary action and rapid responses to mechanical perturbation. Yet the choice of cost function, available information, and assumed dynamics changes model predictions, so apparent optimality must always be evaluated against explicit alternatives.
Feedback and prediction often cooperate within a single movement. A person can prepare an expected grip force before lifting and then revise that force rapidly when tactile receptors signal slip. A reaching movement can begin with a planned trajectory and be corrected using visually or proprioceptively estimated error. These examples illustrate temporal division of labor rather than exclusive selection of a feedforward or feedback regime. Johansson & Westling (1984) documents sensory control in dexterous manipulation, while Wolpert et al. (2011) distinguishes predictive, reactive, and biomechanical components of skilled action. Experimental conclusions must identify the actual measurement interval and the time at which feedback could have causally affected the behavior.
State-Space Models and Learning Rules
A state-space model describes how a latent adaptation state evolves across trials. A simple formulation allows a fraction of previous learning to be retained and adds a change proportional to a trial's error; observed movement then depends on the estimated state and any task-specific strategies. Such models offer interpretable parameters for sensitivity to error and retention, but these parameters are functions of the assumed observation equation and experimental design. Smith et al. (2006) demonstrated how interacting fast and slow processes could reproduce adaptation, forgetting, savings, and spontaneous recovery in reaching. The broader lesson is that superficially similar learning curves may arise from multiple interacting processes rather than a single homogeneous memory.
Error sensitivity is often estimated from trial-by-trial movement changes after controlled perturbation. This strategy requires knowing which error was relevant to the learner. Sensory prediction, target, reward, and agency-related discrepancies can covary in ordinary tasks. Rotated visual feedback is particularly useful because the experimenter can manipulate displayed cursor direction without physically moving the hand along that path. Tseng et al. (2007) experimentally probed the dependence of cerebellar adaptation on sensory prediction errors. More recent clamped feedback methods by Morehead et al. (2017) further separated a visually imposed error from ordinary task performance, showing that implicit adjustments can continue even when the task instruction is to disregard the cursor.
Model identification is a methodological concern. Two-state models are economical descriptions of some learning signatures, but fitted fast and slow components are not automatically proof of two anatomically discrete systems. Parameter recovery depends on sufficient perturbation variation, appropriate baselines, noise modeling, and independent predictions. Smith et al. (2006) offers a powerful example because a spontaneous-recovery test followed from the mathematical model. The strongest modern work combines such out-of-sample tests with manipulations of awareness, feedback, and retention. Learning parameters should be interpreted as properties of the task-model system unless independently supported by evidence about mechanisms and biological substrates.
Bayesian State Estimation and the Role of Priors
Bayesian accounts represent state estimation as an inferential process combining sensory likelihoods with expectations acquired from past experience or task structure. Where observations are uncertain, a prior can influence the estimate more strongly; where observations are precise, their influence may grow. In Körding & Wolpert (2004), sensorimotor learning behavior was examined as a problem of combining prior distributions with uncertain evidence. The formulation is particularly useful for explaining systematic biases as rational consequences of specified informational constraints. Its explanatory power depends on measuring the relevant uncertainty and the prior rather than simply labeling every biased response as Bayesian.
An important distinction separates combining two sensory measurements from inferring the cause of their disagreement. Visual and proprioceptive cues might refer to the same hand, or a displayed cursor may have become a displaced consequence of an action. The brain could down-weight a discrepant cue, infer a changed mapping, or learn to treat two signals as having separate sources. A weighted-average rule cannot capture every such possibility. Sober & Sabes (2003) shows that integration can depend on the motor computation, whereas Ernst & Banks (2002) gives a strong, constrained example of near-optimal visual-haptic combination. Both findings motivate models that state what variables are being inferred, under which causal assumptions.
Bayesian language can obscure biological and psychological questions when a mathematical fit is presented as direct evidence of an organism consciously calculating probabilities. These models are formal characterizations of behavior; their neural realization and representational format require separate study. The strongest tests change cue variance, reliability, temporal order, or prior frequencies and then predict changes in behavior. Franklin & Wolpert (2011) places Bayesian decision theory alongside control and learning mechanisms, and Wolpert et al. (2011) explains why information gathering is itself task dependent. In this synthesis, probabilistic models are evaluated by explanatory specificity, predictive accuracy, and their ability to survive experimentally altered informational conditions.
Optimality, Impedance, and Biomechanical Constraints
An action is produced by a body with mass, inertia, joint geometry, muscle mechanics, and varying interaction forces. A purely kinematic description of the hand cannot by itself specify which muscle forces should be produced. Impedance control modifies the body's mechanical resistance to disturbance through co-contraction and related mechanisms, while optimal feedback models choose control policies according to task costs. Franklin & Wolpert (2011) treats both as complementary computational responses to uncertainty and disturbance. A movement may remain stable because of neural prediction, rapid correction, changed mechanical stiffness, or their combination. Experiments must measure or manipulate these alternatives to identify what has actually changed.
Optimality is relative to the stated objective. Minimizing variability in endpoint position, energy expenditure, time, risk of failure, or physical effort can imply different actions. An apparent deviation from one investigator's preferred cost may be compatible with another cost or with limits on sensing and computation. Scott (2004) emphasizes that understanding natural movement requires integrating limb mechanics and neural control. Because the experimenter usually does not know the organism's complete objective function, observed accuracy cannot be treated as universal evidence of optimization. The most informative tests ask whether behavior changes in a predicted way when the costs and tolerances of the task are manipulated.
Biomechanical constraints also matter when extending sensorimotor theory to prosthetic or robotic systems. A robot arm can implement an accurate state estimator but fail at contact because its actuator compliance differs from a human hand; a human user may compensate for that difference through predictive anticipation and feedback learning. The transfer of control principles across biological and engineered systems therefore depends on specifying the plant, sensors, delay, task, and controller. Miall & Wolpert (1996) and Franklin & Wolpert (2011) provide useful computational starting points, while behavioral validation must establish whether the proposed model actually predicts successful action in the new physical environment.
Reaching under Visuomotor Rotations and Prism Shifts
Visuomotor-rotation paradigms separate physical hand motion from the observed cursor or target relationship. Participants may be asked to reach to targets while a displayed cursor is rotated relative to the hand. Initial endpoint error and subsequent compensatory movements provide a measurable trajectory of adaptation. When feedback returns to its previous mapping, aftereffects can reveal persistent recalibration even after the participant no longer needs to compensate. Krakauer et al. (2005) studied consolidation, interference, and forgetting in rotation learning and showed why experimental sequence and intervening washout can alter conclusions about memory. These designs became foundational because they create an error signal without permanently changing the participant's anatomy.
The same paradigm supports distinct interpretations. A participant might explicitly aim away from a target to offset a known rotation, gradually recalibrate an implicit mapping, or do both. Cursor feedback may signal an unwanted endpoint, an unexpected sensory consequence, or both simultaneously. Mazzoni & Krakauer (2006) documented adaptation that conflicted with an instructed compensatory strategy, while Taylor et al. (2014) used reported aiming direction to distinguish explicit and implicit contributions to performance. The measurement of the aiming report is itself part of the experimental procedure and can affect behavior; it should therefore be described alongside observed movement angles and error magnitudes.
Aftereffects, savings, and transfer require precise definitions. An aftereffect is a systematic movement deviation observed when a perturbation is removed; savings denotes faster subsequent reacquisition under a suitable comparison; transfer concerns generalization to untrained target directions, effectors, or contexts. The existence of one does not guarantee the others. Relearning designs must control for proactive interference, forgotten strategies, baseline drift, and differing movement distributions. Krakauer et al. (2005) illustrates the dependence of consolidation inference on task order, and Smith et al. (2006) shows that different retention timescales can generate behavior that superficially resembles multiple memory phenomena. The scientific question concerns which process changed and what measure isolates it.
Force-Field Perturbations and Dynamic Adaptation
Robotic manipulanda allow experimenters to impose velocity-dependent or position-dependent forces during reaching. A participant who initially moves along a straight path may be deflected when the environment supplies an unfamiliar force. Repeated experience often reduces the perturbation's behavioral impact, and removing the field can produce movement in the opposite direction. The landmark experiments by Shadmehr & Mussa-Ivaldi (1994) established this paradigm for examining adaptive representations of movement dynamics. It makes the body-environment relation a controlled variable and supports direct tests of whether learned compensation depends on direction, posture, or previous exposure.
Improved performance within a force field is not synonymous with one specific internal model. Participants could alter muscle co-contraction, anticipate the expected field, change movement speed, or employ feedback correction more efficiently. Force-channel trials can constrain movement to a trajectory and reveal the predictive forces participants exert, which helps distinguish compensatory feedforward output from simple error reduction. Even then, inferred learning is conditional on the mechanical properties of the apparatus and the assumptions used to estimate forces. Franklin & Wolpert (2011) reviews the multiple computational solutions to disturbance, and Shadmehr et al. (2010) explains how prediction errors can support recalibration.
A central methodological benefit of force-field designs is that they make the cost of inaccurate prediction tangible: unexpected forces displace the hand and alter proprioceptive feedback. This creates opportunities to relate sensory error, force compensation, and movement kinematics across trials. However, the dynamics of the physical limb and arm posture can vary across people, and laboratory field compensation cannot be assumed to generalize to every natural tool-use skill. The distinction between adaptation of dynamics and acquisition of an entirely new skill is explicit in Krakauer et al. (2019). Experimental protocols should therefore report whether the dependent measure captures altered dynamics, increased robustness, changed strategy, or transferable task competence.
The Separation of Implicit and Explicit Adaptation
Implicit recalibration is commonly operationalized as movement change that persists without an instructed aim correction or remains after explicit strategy reports are subtracted. Explicit adaptation is frequently assessed through verbal or spatial reports of where a participant intends to aim. Such measures are informative but not perfectly process-pure. Taylor et al. (2014) used trialwise aiming reports to reveal distinct contributions to visuomotor learning. Mazzoni & Krakauer (2006) showed that a cognitive strategy can initially correct performance while a separate adaptive process continues to change movements. These findings make awareness, strategy, and movement calibration empirically distinguishable dimensions rather than interchangeable labels.
Task-irrelevant clamped feedback creates a particularly sharp dissociation. In Morehead et al. (2017), participants received cursor feedback whose direction was independent of their actual reach and were instructed to ignore that information. Nevertheless, reaching gradually changed, consistent with an automatic component of adaptation driven by sensory prediction discrepancies. The observed adjustment is informative precisely because a successful target-directed strategy should not require responding to the clamp. Its magnitude and saturation, however, are properties of a specified paradigm; it is unwarranted to equate the entire adaptation response with one neural mechanism without further evidence. No-feedback trials and carefully chosen probes remain important for separating direct feedback effects from lasting change.
Recent work challenges the convenience of adding an implicit component and an explicit component as if their contributions were always independent and numerically exhaustive. Chen & Taylor (2026) tested the additivity assumption while accounting for plan-based spatial generalization and temporal decay. The reported results favored partial compensatory relations while failing strict summation. This does not erase the evidence for distinct processes; it means their interaction and the measurement procedure affect decomposition. A strong account identifies how aiming changes the reference frame for recalibration and whether adaptation estimates depend on what target, planned direction, or movement endpoint is used to calculate them. The field's current theoretical disagreement is substantive and experimentally tractable.
Sensory Prediction Error, Performance Error, and Feedback Timing
A sensory prediction error is a difference between expected and observed sensory consequences of action. A task performance error is a difference between an intended outcome and what occurred according to the task criterion. A reach may produce a cursor displaced from its predicted trajectory while still arriving at a rewarded location, or may fail to meet a target despite producing expected feedback. This separation is critical for identifying learning mechanisms. The cerebellum-focused study by Tseng et al. (2007) directly investigated sensory prediction errors during reaching adaptation, and Morehead et al. (2017) used visual clamping to create sensory discrepancy independently of ordinary task success.
Temporal manipulation provides an additional test because feedback influences current control and subsequent learning through partly different pathways. Continuous cursor feedback supplies information during motion; endpoint-only feedback supplies information after the limb has stopped. Wang et al. (2024) advanced the timing of endpoint feedback independently from actual hand position and found that the strength of adaptation depended nonlinearly on when information was presented. This result suggests that the temporal relationship to action planning and predicted consequences matters, rather than a simple assumption that earlier feedback is always better. The exact experimental interpretation depends on feedback presentation, trial timing, and the behavioral measure used.
Investigators should measure more than endpoint error when testing these hypotheses. Adaptation might be expressed as movement-angle change across trials, current-trial corrections, residual error under no-feedback testing, or changes in perceived hand location. These can dissociate. It is also necessary to distinguish a sensory observation that arrives too late to influence the current movement from one that still updates the next movement's control policy. Shadmehr et al. (2010) and Wang et al. (2024) together show that sensory information has a temporal function within control as well as a statistical reliability. Consequently a comprehensive theory must represent uncertainty, causal attribution, and delay as interacting variables.
Dexterous Manipulation and Anticipatory Grip
Precision grip is a useful model of sensorimotor integration because success requires the timing of multiple forces relative to surface friction and object load. In the experiments of Johansson & Westling (1984), tactile afferents and sensorimotor memory supported automatic regulation while objects were lifted, positioned, and replaced. Westling & Johansson (1984) manipulated object weight and surface properties between trials while visual appearance remained similar. The resulting performance illustrates how a movement can draw on memory of prior conditions before new tactile evidence is available. Grip-load coupling is an observable pattern, while the specific predictive computations producing it require further experimental identification.
Online feedback becomes decisive when an object begins to slip or when load differs from expectation. Cutaneous input can signal contact transitions and impending movement relative to the skin; subsequent corrections may stabilize the object before a deliberate conscious adjustment would be feasible. Yet a person also anticipates load based on previous lifts, object appearance, or learned tool dynamics. These mechanisms operate at different moments in one action sequence. Johansson & Westling (1984) provides a quantitative starting point, whereas Wolpert et al. (2011) places predictive manipulation within the larger organization of information extraction and motor decision making. Everyday object handling is therefore an especially useful domain for connecting laboratory computations with ecologically meaningful behavior.
The evidentiary challenge is to distinguish predictive force preparation from reactive adjustments when both alter the same grip-force trace. Experiments can unexpectedly change friction, vary object weight, remove reliable visual cues, or measure force before and after the moment feedback could arrive. Such controls help infer whether previous experience shaped action in advance. However, observations in a two-digit precision grip cannot automatically represent complex whole-hand manipulation or interactions with deformable tools. When extending these findings to rehabilitation or engineered hands, the mechanical range, tactile resolution, and information delays of the system must be measured. Franklin & Wolpert (2011) explains why controller mechanics and sensory uncertainty jointly shape action.
Cerebellar Prediction and Neural Circuit Evidence
The cerebellum is central to many experimental interpretations of adaptation because anatomical and lesion evidence links its circuits to calibration of movement-relevant predictions. A precise claim, however, concerns the measured function: changing movement after an unexpected sensory consequence is a behavioral signature, while an attribution to a particular circuit requires anatomical, physiological, or intervention evidence. Tseng et al. (2007) manipulated conditions relevant to sensory prediction error and reported that cerebellar-dependent adaptation required prediction of sensory movement consequences. This result supports a mechanistic account while leaving open the question of how other networks cooperate in prediction, perception, and the formation of conscious movement strategies.
Modern circuit experiments reveal how cerebellar processing can represent both self-generated and externally caused sensory changes. Zobeiri & Cullen (2024) investigated mechanisms through which cerebellar pathways encode sensory prediction errors during active movement. The interpretation is important because sensory input arising from movement must be compared with an internally derived expectation on an appropriate temporal scale. Such findings strengthen links between behavioral theories and biological circuits, but the species, behavioral tasks, recorded cell populations, and forms of manipulation constrain their generality. A circuit-level signal that correlates with a mismatch does not alone demonstrate that this signal is sufficient to generate all forms of motor adaptation.
Cerebellar explanations should also be connected to the specific perturbation. Visuomotor rotations, altered dynamics, tactile surprises, and vestibular reafference involve different receptors and changing relations among body, task, and world. Shadmehr et al. (2010) presents the broader adaptation framework, while Wolpert et al. (2011) places neural prediction within multiple forms of action control. The promising program is to specify which predictions are compared with which observations, how errors influence future commands, and whether independent circuit manipulations produce the predicted dissociation across tasks. The result would connect an algorithmic explanation to a biological mechanism without treating a single structure as a complete description of movement.
Cortical, Spinal, and Long-Latency Feedback Responses
Sensorimotor feedback reaches several interacting control levels. Peripheral reflex pathways can generate short-latency responses to stretch, whereas longer-latency responses can incorporate more elaborate information about task context and intended action. Pruszynski & Scott (2012) synthesized evidence for rapid sensorimotor processing and its relationship to optimal feedback control. The core observation is that corrections occurring before ordinary deliberate reaction times can still be sensitive to the goal and mechanical relevance of a perturbation. The presence of such task-dependent feedback questions the assumption that fast responses must reflect context-free reflex arcs.
Motor cortical and related circuits participate in the adjustment of activity when unexpected mechanical forces displace a limb. Scott (2004) argued that optimal feedback control provides a coherent framework for interpreting voluntary movement and neural response to disturbances. It is essential, however, to distinguish a theoretical control policy from proof that individual neural populations literally encode a mathematically optimal policy. Recording, stimulation, and lesion methods have different inferential powers. Behavioral correction latency can limit the set of feasible mechanisms, whereas direct neural recordings and interventions are needed to identify the pathway that mediates the correction.
Biomechanical properties operate together with neural responses. A stiffened limb may resist displacement without requiring a large delayed correction; a more compliant movement may depend more visibly on active feedback. This coupling complicates simple classifications of responses as either spinal or cortical. Franklin & Wolpert (2011) emphasizes both feedback control and impedance, and Pruszynski & Scott (2012) reviews how long-latency responses can be arranged according to task goals. A well-designed experiment therefore records forces, kinematics, temporal changes in muscle activation, and environmental load. Such multimodal measurement supports a layered explanation in which the physical body and distributed neural control cooperate to stabilize action.
Agency, Self-Produced Sensation, and Sensory Attenuation
Actions normally produce predictable sensations. One hypothesis is that the nervous system uses a copy of an outgoing motor command to forecast these sensations, adjusting the interpretation of input that matches the prediction. Blakemore et al. (1999) examined perceptual consequences of self-produced tactile stimulation and linked sensory attenuation to predictive mechanisms. The finding is important for motor psychology because the sensory stream depends partly on what the organism itself has done. Yet perceptual attenuation, motor correction, and the conscious sense of being the agent of an action are different dependent variables, and they require different behavioral tests.
The conceptual history of self-produced sensation extends to the older physiological literature on reafference and corollary discharge. Sperry (1950) studied adaptive visual control under experimentally altered visual conditions. A later explanatory review by Wolpert & Flanagan (2001) linked motor prediction to judgments about whether sensations were self-generated. The mechanism remains a scientific hypothesis that can be assessed through changes in temporal synchrony, congruence, predictability, and voluntary control. The central question is what mismatch changes a person's response, not whether one philosophical interpretation of agency follows from it.
Agency judgments also depend on higher-level expectations and contextual cues, and a discrepancy between prediction and sensation may be interpreted differently depending on the task. The attenuation of sensation during self-generated movement should therefore not be used as a standalone diagnostic test of selfhood or consciousness. Blakemore et al. (1999) supports a constrained account of predicted tactile consequences, while Wolpert et al. (2011) situates these comparisons among multiple mechanisms of skilled sensorimotor behavior. In technologically mediated action, feedback from a cursor, prosthesis, or robot may become predictable through learning without establishing that the device is represented in the same way as a biological limb. That claim would require carefully designed measures of agency, body representation, and adaptation.
Rehabilitation, Assistive Devices, and Tool-Mediated Action
Sensorimotor rehabilitation is informed by experiments on state estimation, sensory reliability, feedback scheduling, and retention of learned mappings. A person controlling a prosthesis or using assisted movement may need to coordinate unfamiliar sensors, a new physical plant, and delayed visual or tactile feedback. Franklin & Wolpert (2011) describes computational concepts relevant to this challenge, while Krakauer et al. (2019) distinguishes controlled adaptation from the broader development of functional skill. Translation requires measuring whether a laboratory change improves meaningful activity under varied environments. A reduced adaptation error on a reaching task is evidence of learning within that task, not automatically of improved daily functioning.
Technological support can alter both what the user sees and how the task can be controlled. When visual feedback is amplified, delayed, or replaced with another sensory code, sensorimotor estimates may recalibrate, but users can also adopt explicit strategies that increase performance without changing the underlying mapping. Taylor et al. (2014) and Morehead et al. (2017) make this distinction experimentally salient. Evaluation of an interface should specify its latency, the user's available sensory channels, actuator dynamics, training schedule, and task-specific success criteria. The same apparatus can support different control strategies at different stages of practice.
Clinical translation further depends on the population studied and the safety of the intervention. Findings from healthy young adults may not apply to people with sensory neuropathy, movement disorders, stroke, or multiple coexisting impairments. Proske & Gandevia (2012) provides physiological context for proprioceptive differences, and Scott (2004) explains why limb mechanics and feedback mechanisms must be considered jointly. This monograph does not prescribe rehabilitation protocols. It identifies the experimental variables and inferential gaps that a qualified clinical research program would need to address: mechanism-specific outcomes, controlled comparisons, individualized measurement, meaningful function, longer-term retention, and adverse effects.
Behavioral, Psychophysical, and Physiological Measurement
Measurement of sensorimotor integration begins by separating what the investigator controls from what is inferred. A visuomotor-rotation task controls the displayed mapping between movement and feedback; a force-field task controls a mechanical perturbation; a visual-haptic discrimination task controls the available sensory estimates. Dependent measures can include endpoint error, reaching angle, correction latency, grip-force coordination, perceptual judgments, or post-exposure aftereffects. These observations answer different questions. Ernst & Banks (2002) examined cue combination through perceptual discrimination, whereas Shadmehr & Mussa-Ivaldi (1994) studied changes in reaching under altered dynamics. Treating these measures as interchangeable obscures the mechanisms responsible for behavior. A study should name the task variable, its timing, units, baseline, and intended psychological construct.
Experimental controls are especially important when participants can consciously alter their behavior. Asking where a person aimed provides evidence concerning a reportable strategy, while measuring movements without visual feedback can probe persistent adaptation. Taylor et al. (2014) illustrates an explicit-report method, and Morehead et al. (2017) demonstrates the use of task-irrelevant clamped feedback. Neither method offers a universally process-pure readout: reporting may alter performance, and a no-feedback trial may still express prior strategic preparation. Inferences therefore improve when complementary measures converge under controlled manipulations. Investigators should report exclusion rules, trial numbers, the distribution of target directions, and whether apparent learning is measured within trials or across trials.
Physiological measures add another layer. Muscle activity can help distinguish preplanned force from corrective response; limb kinematics can reveal mechanical effects; perceptual matching can characterize proprioceptive state estimation; neural recordings can test hypotheses about candidate circuits. Pruszynski & Scott (2012) explains the relevance of response latency and task-dependent correction, while Proske & Gandevia (2012) clarifies distinct dimensions of proprioceptive measurement. The most informative designs coordinate these signals in time. A correction recorded after a mechanical perturbation cannot be attributed to feedback that had not yet reached the relevant controller. Such temporal constraints turn a descriptive movement trace into evidence capable of adjudicating among causal accounts.
Model Fitting, Generalization, and Reproducibility
Computational models are valuable when they make sufficiently precise predictions to be falsified. State-space models predict trial-to-trial changes from retention and error sensitivity; Bayesian models predict responses to manipulation of cue reliability and prior expectations; feedback-control models predict corrections under specified costs and dynamics. Smith et al. (2006) showed how a two-rate model yields testable patterns such as spontaneous recovery, while Körding & Wolpert (2004) studied probabilistic structure in sensorimotor inference. A model is supported more strongly when it predicts behavior in an independently manipulated condition than when it merely reproduces observations used for parameter fitting. Comparing multiple candidate models and checking parameter recoverability are necessary steps in interpreting fitted coefficients.
Generalization tests ask whether a recalibrated controller transfers across movement direction, workspace, effector, task objective, or sensory modality. Transfer depends on what representation the model assumes and on what practice was actually experienced. Krakauer et al. (2005) demonstrated that interference and consolidation conclusions depend on training and testing sequence. Krakauer et al. (2019) discusses the broader relation between laboratory motor learning and acquisition of complex skills. Repeated practice with one cursor rotation may give a stable estimate of local adaptation while saying little about navigating unfamiliar environments or using an assistive device in daily life. A rigorous claim of transfer must specify both the source and target conditions and compare performance with appropriate controls.
Reproducibility also concerns analytic decisions. Researchers may choose how to remove outlier trials, normalize baseline performance, separate explicit from implicit components, fit retention parameters, or define asymptotic adaptation. These choices can change results even with the same raw trajectories. Chen & Taylor (2026) revisited additivity assumptions in the interpretation of adaptation components; Tsay et al. (2024) reviewed a broader framework of reasoning, refinement, and retrieval in motor learning, emphasizing that interpretation of adaptation data depends on task structure, strategy, and the process being measured. Transparent trial-level data, preregistered contrasts when appropriate, explicit model code, and sensitivity analyses allow independent laboratories to determine whether a conclusion follows from the evidence or from one analytic convention.
Critical Appraisal of Competing Accounts
The first major explanatory debate concerns how much motor behavior should be attributed to a stored predictive model versus flexible feedback policies and bodily mechanics. Predictive accounts explain anticipatory compensation and the updating of movements after unexpected sensory consequences Wolpert et al. (1995) Shadmehr et al. (2010). Feedback-control accounts explain rapid, goal-dependent correction during movement Scott (2004) Pruszynski & Scott (2012). Impedance control explains how the physical plant can resist perturbation without waiting for a new sensory correction Franklin & Wolpert (2011). These are partly complementary mechanisms, and a task can recruit more than one. A discriminating experiment should alter feedback availability, perturbation timing, or mechanical stiffness and then test which account predicts the resulting response.
A second debate concerns the decomposition of adaptation into implicit and explicit processes. The work of Mazzoni & Krakauer (2006) and Taylor et al. (2014) established strong experimental reasons to distinguish automatic recalibration from strategic re-aiming. Clamped cursor feedback further showed that behavior can change in response to task-irrelevant discrepancies Morehead et al. (2017). Yet fitting a total movement change as the arithmetic sum of two isolated processes assumes independence that may fail when aiming changes the reference coordinates for learning. Chen & Taylor (2026) offers recent evidence against strict additivity under the tested conditions. Thus the scientific advance lies in specifying the interaction, not merely deciding whether one named process dominates all experiments.
A third debate concerns whether apparently optimal integration implies general-purpose Bayesian computation. Ernst & Banks (2002) provides a compelling case of near-optimal visual-haptic integration in a constrained discrimination task, while Sober & Sabes (2003) demonstrates that the use of visual and proprioceptive information can differ among motor computations. The generalization from a favorable model fit to a comprehensive theory of movement is therefore a substantive empirical hypothesis. Researchers must ask what prior knowledge is available, whether sensory errors share a cause, how costs of failure vary, and whether the participant's strategy is stable across time. The evidence supports an increasingly differentiated account of sensorimotor organization with task-specific state estimation, prediction, mechanics, and learning.
Embodied Artificial Systems and Human–Artificial Sensorimotor Relations
The Artificial Era makes sensorimotor integration a comparative scientific topic because artificial controllers can possess sensors, actuator commands, state estimators, feedback delays, and models trained from interaction. A robotic system manipulating an object confronts formal problems akin to those studied in human motor control: estimating an unobserved state, anticipating the effects of action, and correcting after delayed evidence. Franklin & Wolpert (2011) offers the computational concepts needed for a controlled comparison. To evaluate a specific artificial agent, one must operationalize its sensory stream, internal state, actuator dynamics, policy updates, and task performance. Similar mathematical descriptions across systems establish functional comparability for particular mechanisms; they do not independently establish identical biological implementation or subjective experience.
Human–artificial cooperation introduces a coupled control loop. A person operating a telepresence device, robot, or prosthetic interface receives feedback shaped by a second system, while commands from the person alter that system's next state. The quality of coordination depends on delays, alignment between interface and action, the reliability of sensory channels, learned task mappings, and the distribution of control responsibilities. Shadmehr et al. (2010) and Wolpert et al. (2011) describe the broader sensorimotor mechanisms relevant to these challenges. Empirical research should distinguish performance improved by algorithmic assistance from changes in the human user's own predictive calibration. Both can produce better endpoint accuracy while having different implications for learning, agency, and the ability to operate without assistance.
In the Aisentica intellectual horizon, Artificial designates a historical category for investigating nonbiological systems and their relations with Homo. Here that horizon contributes research questions rather than substituting a philosophical proposition for an experiment. The observable issues are whether state estimation can be measured, whether a controller learns from sensory prediction errors, whether two interacting agents coordinate action, and how the resulting behavior generalizes to novel contexts. Krakauer et al. (2019) provides a reason to distinguish adaptation from skill; Blakemore et al. (1999) reminds us that human judgments about self-produced sensation require specific perceptual measures. Such distinctions support comparative research that connects psychological constructs to reproducible observation and explicit models.
Conclusion
Sensorimotor integration is the experimentally investigable coordination of sensation, inferred body and environmental state, movement preparation, online feedback, and adaptation. Its most robust evidence comes from converging paradigms: vision and proprioception contribute to state estimation; haptic and visual information can be weighted according to task-specific reliability; force-field and visuomotor perturbations reveal adaptation; and rapid feedback responses depend on task goals and bodily mechanics. The historical work of Held & Hein (1963), Shadmehr & Mussa-Ivaldi (1994), and Wolpert et al. (1995) established complementary foundations, while later research made experimental distinctions sharper. Scientific progress has depended on increasingly precise definitions of what changes within a movement, after a trial, and across extended practice.
The synthesis identifies three recurrent methodological obligations. First, an observed movement deviation needs an explicit account of which sensory, performance, or motivational error is being measured. Second, computational models require independent tests that separate descriptions from candidate biological mechanisms. Third, conclusions about learning must distinguish strategy, recalibration, retention, and transfer. Taylor et al. (2014), Morehead et al. (2017), and Chen & Taylor (2026) illustrate why familiar measures of adaptation cannot always be decomposed by simple addition. Recent feedback-timing results Wang et al. (2024) and circuit studies Zobeiri & Cullen (2024) provide further opportunities to connect temporally specific behavior to mechanisms. No single paradigm exhausts the organization of sensory and motor coordination.
Future research will benefit from studies that manipulate several sensory uncertainties independently, compare models on held-out conditions, combine behavioral and neural interventions, and establish how learning generalizes across contexts and populations. Better longitudinal designs can clarify how sensorimotor calibration develops and changes with experience, while studies of human–artificial systems can test the effects of delays, distributed control, and adaptive feedback. The enduring scientific conclusion is that reliable action emerges from interacting estimators, predictions, control policies, mechanical constraints, and learning processes whose contributions can be experimentally separated and recombined. The goal of this research dossier has been to organize the strongest established evidence and the remaining questions into one coherent account Wolpert et al. (2011) Krakauer et al. (2019).
References
- Blakemore, S.-J., Frith, C. D., & Wolpert, D. M. (1999). Spatio-temporal prediction modulates the perception of self-produced stimuli. Journal of Cognitive Neuroscience, 11(5), 551–559. https://doi.org/10.1162/089892999563607
- Chen, Y., & Taylor, J. A. (2026). Revisiting the explicit-implicit additivity assumption in visuomotor adaptation. Journal of Neurophysiology, 135(5), 1298–1314. https://doi.org/10.1152/jn.00525.2025
- Cisneros, E., Karny, S., Ivry, R. B., & Tsay, J. S. (2026). A systematic investigation reveals dissociable effects of ageing on implicit and explicit components of sensorimotor learning. Nature Human Behaviour, 10(8), 1510–1530. https://doi.org/10.1038/s41562-026-02468-7
- Ernst, M. O., & Banks, M. S. (2002). Humans integrate visual and haptic information in a statistically optimal fashion. Nature, 415, 429–433. https://doi.org/10.1038/415429a
- Franklin, D. W., & Wolpert, D. M. (2011). Computational mechanisms of sensorimotor control. Neuron, 72(3), 425–442. https://doi.org/10.1016/j.neuron.2011.10.006
- Hein, A., & Held, R. (1967). Dissociation of the visual placing response into elicited and guided components. Science, 158(3799), 390–392. https://doi.org/10.1126/science.158.3799.390
- Held, R., & Hein, A. (1963). Movement-produced stimulation in the development of visually guided behavior. Journal of Comparative and Physiological Psychology, 56, 872–876. https://doi.org/10.1037/h0040546
- Johansson, R. S., & Westling, G. (1984). Roles of glabrous skin receptors and sensorimotor memory in automatic control of precision grip when lifting rougher or more slippery objects. Experimental Brain Research, 56(3), 550–564. https://doi.org/10.1007/BF00237997
- Körding, K. P., & Wolpert, D. M. (2004). Bayesian integration in sensorimotor learning. Nature, 427, 244–247. https://doi.org/10.1038/nature02169
- Krakauer, J. W., Ghez, C., & Ghilardi, M. F. (2005). Adaptation to visuomotor transformations: Consolidation, interference, and forgetting. Journal of Neuroscience, 25(2), 473–478. https://doi.org/10.1523/JNEUROSCI.4218-04.2005
- Krakauer, J. W., Hadjiosif, A. M., Xu, J., Wong, A. L., & Haith, A. M. (2019). Motor learning. Comprehensive Physiology, 9(2), 613–663. https://doi.org/10.1002/cphy.c170043
- Mazzoni, P., & Krakauer, J. W. (2006). An implicit plan overrides an explicit strategy during visuomotor adaptation. Journal of Neuroscience, 26(14), 3642–3645. https://doi.org/10.1523/JNEUROSCI.5317-05.2006
- Miall, R. C., & Wolpert, D. M. (1996). Forward models for physiological motor control. Neural Networks, 9(8), 1265–1279. https://doi.org/10.1016/S0893-6080(96)00035-4
- Morehead, J. R., Taylor, J. A., Parvin, D. E., & Ivry, R. B. (2017). Characteristics of implicit sensorimotor adaptation revealed by task-irrelevant clamped feedback. Journal of Cognitive Neuroscience, 29(6), 1061–1074. https://doi.org/10.1162/jocn_a_01108
- Proske, U., & Gandevia, S. C. (2012). The proprioceptive senses: Their roles in signaling body shape, body position and movement, and muscle force. Physiological Reviews, 92(4), 1651–1697. https://doi.org/10.1152/physrev.00048.2011
- Pruszynski, J. A., & Scott, S. H. (2012). Optimal feedback control and the long-latency stretch response. Experimental Brain Research, 218(3), 341–359. https://doi.org/10.1007/s00221-012-3041-8
- Scott, S. H. (2004). Optimal feedback control and the neural basis of volitional motor control. Nature Reviews Neuroscience, 5, 532–545. https://doi.org/10.1038/nrn1427
- Shadmehr, R., & Mussa-Ivaldi, F. A. (1994). Adaptive representation of dynamics during learning of a motor task. Journal of Neuroscience, 14(5), 3208–3224. https://doi.org/10.1523/JNEUROSCI.14-05-03208.1994
- Shadmehr, R., Smith, M. A., & Krakauer, J. W. (2010). Error correction, sensory prediction, and adaptation in motor control. Annual Review of Neuroscience, 33, 89–108. https://doi.org/10.1146/annurev-neuro-060909-153135
- Smith, M. A., Ghazizadeh, A., & Shadmehr, R. (2006). Interacting adaptive processes with different timescales underlie short-term motor learning. PLOS Biology, 4(6), e179. https://doi.org/10.1371/journal.pbio.0040179
- Sober, S. J., & Sabes, P. N. (2003). Multisensory integration during motor planning. Journal of Neuroscience, 23(18), 6982–6992. https://doi.org/10.1523/JNEUROSCI.23-18-06982.2003
- Sperry, R. W. (1950). Neural basis of the spontaneous optokinetic response produced by visual inversion. Journal of Comparative and Physiological Psychology, 43(6), 482–489. https://doi.org/10.1037/h0055479
- Taylor, J. A., Krakauer, J. W., & Ivry, R. B. (2014). Explicit and implicit contributions to learning in a sensorimotor adaptation task. Journal of Neuroscience, 34(8), 3023–3032. https://doi.org/10.1523/JNEUROSCI.3619-13.2014
- Tsay, J. S., Kim, H. E., McDougle, S. D., Taylor, J. A., Haith, A., Avraham, G., Krakauer, J. W., Collins, A. G. E., & Ivry, R. B. (2024). Fundamental processes in sensorimotor learning: Reasoning, refinement, and retrieval. eLife, 13, e91839. https://doi.org/10.7554/eLife.91839
- Tseng, Y.-W., Diedrichsen, J., Krakauer, J. W., Shadmehr, R., & Bastian, A. J. (2007). Sensory prediction errors drive cerebellum-dependent adaptation of reaching. Journal of Neurophysiology, 98(1), 54–62. https://doi.org/10.1152/jn.00266.2007
- Wang, T., Avraham, G., Tsay, J. S., Thummala, T., & Ivry, R. B. (2024). Advanced feedback enhances sensorimotor adaptation. Current Biology, 34(5), 1076–1085.e5. https://doi.org/10.1016/j.cub.2024.01.073
- Westling, G., & Johansson, R. S. (1984). Factors influencing the force control during precision grip. Experimental Brain Research, 53(2), 277–284. https://doi.org/10.1007/BF00238156
- Wolpert, D. M., Ghahramani, Z., & Jordan, M. I. (1995). An internal model for sensorimotor integration. Science, 269(5232), 1880–1882. https://doi.org/10.1126/science.7569931
- Wolpert, D. M., & Flanagan, J. R. (2001). Motor prediction. Current Biology, 11(18), R729–R732. https://doi.org/10.1016/S0960-9822(01)00432-8
- Wolpert, D. M., Diedrichsen, J., & Flanagan, J. R. (2011). Principles of sensorimotor learning. Nature Reviews Neuroscience, 12, 739–751. https://doi.org/10.1038/nrn3112
- Zobeiri, O. A., & Cullen, K. E. (2024). Cerebellar Purkinje cells combine sensory and motor information to predict the sensory consequences of active self-motion in macaques. Nature Communications, 15, Article 4003. https://doi.org/10.1038/s41467-024-48376-0