RESEARCH DOSSIER
Neural Control of Movement and Action Selection
Abstract. Neural control of movement and action selection comprises the distributed processes through which nervous systems represent alternative actions, prepare motor commands, regulate commitment and vigor, generate coordinated muscle activity, and correct unfolding movement. This fundamental critical narrative monograph reconstructs the experimental history of spinal integration, descending motor control, basal ganglia circuits, cerebellar computation, and neural population dynamics. It evaluates primate electrophysiology, human neuroimaging and perturbation studies, optogenetic circuit interventions, movement-disorder observations, computational optimal feedback theories, and brain-computer interface experiments, including original 2025–2026 experiments on basal ganglia movement-specific output, action counting, thalamocortical learning, and incentive-dependent movement initiation. Particular attention is given to evidence that cortical choice specification, basal ganglia modulation, cerebellar prediction, and spinal mechanics are dynamically coupled, and to findings that challenge a single serial selection switch or a simple direct-pathway go/indirect-pathway stop account. The synthesis distinguishes movement initiation, motor execution, vigor, selection, adaptation, and conscious agency; examines cross-species inference and methodological constraints; and identifies testable problems in inter-regional timing and causal circuit interaction. It is a critical scientific synthesis of published literature rather than an original experiment, systematic review, or meta-analysis.
Keywords: neural control of movement; action selection; motor cortex; motor preparation; basal ganglia; direct pathway; indirect pathway; motor vigor; cerebellum; spinal motor circuits; motor synergies; optimal feedback control; motor neuroscience; motor initiation; population dynamics; motor learning; neural perturbation; motor disorders; brain-computer interfaces; cortico-basal ganglia circuits; thalamocortical motor learning
The Scientific Object: From Candidate Action to Controlled Movement
Neural control of movement and action selection concerns how a nervous system transforms available goals, environmental information, and bodily constraints into an executed action. Action selection identifies which behavioral possibility is currently pursued; movement initiation concerns when activity is released into overt output; motor execution concerns time-varying muscle commands and their mechanical consequences. These functions overlap in real circuits but their measures differ. A choice between two targets may be accurate while initiation is delayed, or an action may begin quickly yet follow an imprecise trajectory. Classical basal ganglia circuit accounts and contemporary analyses of initiation motivate this multidimensional definition Alexander et al. (1986) Klaus et al. (2019).
A motor act is both a neural event and a physical interaction. Position, velocity, torque, co-contraction, force, and contact are altered by the body's inertia, limb geometry, muscle mechanics, and external loads. Consequently a neural activity pattern does not have a fixed behavioral meaning independently of task, posture, effectors, and environmental context. A defensible explanation links neural perturbations to observed kinematics and kinetics while specifying which behaviorally relevant variable changes. Hierarchical accounts emphasize that descending signals interact continuously with spinal circuitry and muscle properties rather than issuing a complete trajectory to passive tissue Loeb et al. (1999).
This monograph asks where competing actions are represented, how commitment becomes timely, how cortical activity changes between preparation and execution, how basal ganglia output regulates initiation and vigor, and how cerebellar, spinal, and sensory pathways correct movements. It evaluates electrophysiological correlations, targeted interventions, formal models, and neurological dissociations as different sources of evidence. General reward learning and the development of habitual preferences remain neighboring research objects; here they enter only when they clarify the control of a particular movement. Such boundaries avoid treating selection, learning, and skilled coordination as interchangeable outcomes Dudman & Krakauer (2016) Shadmehr et al. (2010).
Definitions, Levels of Analysis, and Operational Measures
An action candidate is a potential response specified by a task, inferred from behavior, or decoded from a neural population before execution. A motor plan is a preparatory state that is informative about prospective movement, although its causal role must be tested rather than presumed from decodability. Initiation is often operationalized as the onset of electromyographic activity, movement displacement, or a task-defined response; these operationalizations have different latencies and sensitivity. Commitment refers to reducing deliberation in favor of an action, whereas vigor concerns the energetic or temporal intensity of its performance. The distinction matters when neural perturbations change reaction time without changing target choice Thura & Cisek (2017) Klaus et al. (2019).
Behavioral measurement begins with reaction time, selection accuracy, peak speed, endpoint error, trajectory curvature, timing variability, and anticipatory muscle activity. Neurophysiological measurement adds action potentials, local field potentials, calcium-dependent signals, imaging contrasts, and patterned responses to stimulation. Formal models supply explicit quantities such as estimated state, expected cost, control policy, or decision threshold. Each quantity corresponds to a limited class of observations, and apparent agreement across modalities requires explicit mapping. A correlated firing rate, for example, cannot establish that the encoded movement variable is the unique variable controlling that neuron Georgopoulos et al. (1982) Shenoy et al. (2013).
Three explanatory questions must be kept separate: what the system computes, how its circuits implement the computation, and which interventions demonstrate causal necessity or sufficiency. Optimal control can explain why task-irrelevant movement variability is tolerated, yet several neural architectures could implement the same policy. A selective circuit manipulation may establish involvement while leaving its internal algorithm unsettled. Conversely, neural population models may predict recorded firing accurately while missing afferent drive or biological implementation. Evidence strength therefore grows when behavioral predictions, single-unit or population recordings, perturbations, and anatomical connectivity converge under transparent alternative hypotheses Todorov & Jordan (2002) Sauerbrei et al. (2020).
From Reflex Integration to Voluntary Motor Physiology
The early physiology of movement focused on how convergent excitation, inhibition, and reciprocal organization produce coordinated motor responses. Work in the Sherrington tradition established spinal integration as a genuine computational problem, not merely a cable carrying cerebral commands. Subsequent direct study of inhibitory pathways used afferent stimulation and electrophysiology to clarify circuit interactions that shape muscle activation. Eccles, Fatt, and Landgren (1956) provided influential experiments on central inhibition from large afferent fibers, although the historical recording methods cannot resolve the cell-type diversity accessible to current molecular methods Eccles et al. (1956).
The development of chronic recording in behaving primates transformed the study of voluntary movement. Evarts (1968) linked pyramidal tract neuron activity with forces generated during trained movements, making the relation between descending signals and behavioral output experimentally tractable. Such recordings demonstrated that motor cortex activity covaries with aspects of execution under controlled task conditions. They did not imply that each recorded neuron represents a single invariant force variable. Experimental context, sampling of specialized projection neurons, and mechanical coupling all complicate extending a local result to the entirety of voluntary behavior Evarts (1968).
Two subsequent traditions expanded the problem. Single-cell studies asked how preferred movement parameters are distributed across motor cortex; stimulation and lesion experiments asked how neuronal ensembles and subcortical structures alter coordinated behaviors. Meanwhile mathematical models framed movement as control under delay, uncertainty, and redundancy. The modern literature inherits all three approaches and needs them simultaneously. A historically influential label such as 'motor command' can denote a descending signal, a population state, or a model variable; using it without a specified level obscures rather than resolves disagreement Georgopoulos et al. (1982) Bizzi et al. (1991) Todorov & Jordan (2002).
Spinal Integration, Reciprocal Inhibition, and Descending Influence
Spinal circuits transform descending input into muscle-specific activity through motoneurons, inhibitory interneurons, sensory afferents, and local coordination. Reciprocal inhibition can reduce opposing muscle activation during some movements, while co-contraction can stabilize a limb against uncertain loads. Both patterns may be functional depending on the task. Classic afferent studies reveal how circuitry changes the effective impact of peripheral signals on motoneuron recruitment. These experiments establish component mechanisms of movement, even though laboratory spinal preparations and imposed stimulation differ from fully voluntary movement in awake animals and humans Eccles et al. (1956) Loeb et al. (1999).
The spinal cord can contribute organized output beyond a sequence of independent single-muscle reflexes. Bizzi and colleagues described how interacting spinal modules and biomechanics could reduce the effective dimensionality of motor control, helping to explain coordinated patterns. Evidence from microstimulation and muscle recordings is compatible with useful modular organization but does not establish one universally fixed library of movement primitives. The mapping from stimulation to muscle output depends on state, preparation, and circuit accessibility. A successful account must explain both reusable coordination and the flexibility that complex tasks require Bizzi et al. (1991) Overduin et al. (2012).
Descending motor signals operate within this adaptive substrate. Corticospinal projections can influence detailed muscle recruitment, while brainstem pathways provide broad control over posture and locomotion; their exact functional contributions depend on species, movement, and behavioral context. The effect of a cortical intervention can be amplified or buffered by downstream systems, so a movement impairment does not map one-to-one onto the lost computation. Physiologically informative studies combine neural recordings with timing-specific perturbations and carefully characterized movement tasks. This perspective anticipates later evidence that motor cortex recruits short-latency effector pathways selectively according to behavioral context Miri et al. (2017).
Motor Cortical Maps and the Limits of Somatotopy
Mapping motor cortex initially made voluntary control seem topographically straightforward: body parts occupy distinguishable regions and local interventions recruit recognizable muscles. Yet motor maps change with stimulation duration, prior activity, and functional demands. Long-train microstimulation in monkeys can elicit coordinated multijoint behaviors, sometimes converging toward characteristic postures from different starting positions. Graziano, Taylor, and Moore (2002) thereby demonstrated that the behavioral output associated with a stimulation site may span substantial portions of the body rather than a single joint or muscle Graziano et al. (2002).
Directional tuning provides another organizing observation. Georgopoulos and colleagues (1982) recorded motor-cortical neurons while monkeys reached in multiple planar directions, finding systematic relations between discharge and movement direction. Population descriptions can recover directional information more accurately than individual noisy cells. The famous population-vector interpretation was historically productive because it made explicit predictions about jointly distributed activity. Nevertheless, directional tuning may be influenced by correlated posture, force, target, and preparation variables; task-conditioned decoding is evidence of information, not sufficient evidence of a biological decoding mechanism Georgopoulos et al. (1982).
A modern map is therefore a conditional description of causal access and population organization. Functional topography coexists with overlapping representations, recurrent interactions, and many-to-many links between neurons and muscles. The appropriate unit of analysis depends on the question: single-cell physiology for synaptic specificity, circuits for stimulation effects, population dimensions for preparation, and measurable outputs for behavioral control. Competing accounts should specify what would differ under altered posture, feedback, or task rules. Maps retain their empirical value when their scope is defined rather than silently treated as complete movement programs Graziano et al. (2002) Gallego et al. (2017).
Basal Ganglia, Cerebellum, and Cortex as Interacting Systems
The corticospinal system is embedded in loops involving striatum, pallidum, substantia nigra, thalamus, cerebellum, and multiple cortical regions. Parallel basal ganglia–cortical channels motivated influential models of functionally differentiated motor and associative activity. That anatomical organization is an essential foundation for studying action selection, but parallel channels are connected through common behavioral requirements and interacting downstream structures. The basal ganglia have especially prominent roles in modulating action onset, suppression, and vigor, while cerebellar systems contribute prediction, timing, calibration, and broader task preparation Alexander et al. (1986) Bostan & Strick (2018).
Experimental comparisons cannot simply assign selection to one nucleus and correction to another. Evidence for bidirectional pathways between basal ganglia-related and cerebellar-related systems shows a more integrated architecture than an isolated serial chain. The cortico-cerebellar loop can help maintain activity before a movement, and some cerebellar manipulations affect decisions based on sensory evidence. Both findings require causal interpretation in their specific preparations and tasks. They broaden the plausible functions of cerebellar networks without establishing that the same computations occur uniformly across all cerebellar regions Gao et al. (2018) Deverett et al. (2018).
The organizing premise of the following chapters is distributed causal control. Cortical populations represent and transform candidate actions, basal ganglia circuits shape the conditions and intensity of commitment, cerebellar circuits contribute predictive and temporal operations, and spinal networks realize output through constrained bodies. These are hypotheses about interacting roles whose details must be tested, not a diagram of immutable divisions of labor. The evidential task is to identify where activity changes behavior, under what conditions, and through what pathways. Interventions that vary timing and target are more informative than global measures of activation alone Turner & Desmurget (2010) Manto et al. (2012).
Affordances, Competing Reach Plans, and Premotor Choice
An organism confronting several reachable targets may partially specify more than one prospective movement before the decision is settled. Premotor recording experiments provide a direct way to study this intermediate state because each possible reach can be associated with a measurable pattern of activity. In a delayed decision task, Cisek and Kalaska (2005) reported that dorsal premotor cortex reflected multiple potential reach directions before a final direction became dominant. The evidence supports parallel prospective specification under their experimental conditions, and it challenges a strictly serial picture in which a completed abstract decision always precedes the first motor plan Cisek & Kalaska (2005).
Parallel preparation does not imply that every available action is represented with equal strength or that each neural trajectory corresponds to a consciously entertained option. Cues, reward expectations, biomechanical demands, urgency, and attention affect the competition. A neural signal informative about several targets may reflect a mixture of neurons with specialized preferences or a continuously evolving population state; the correct account requires more detailed recording and task perturbations. The behavioral measure of choice must also be separated from the time at which subjects are permitted to move. These qualifications matter when comparing premotor findings with basal ganglia manipulations of commitment Cisek & Kalaska (2005) Thura & Cisek (2017).
A useful experimental distinction concerns the information that defines an option. A target location is not itself an executable movement; a reach also requires specifying an effector, starting posture, force, and route around obstacles. Many apparent choice signals can therefore be conditioned by the geometry of the action space. Experimental designs can vary the mapping between visual targets and motor commands to ask whether activity codes a goal, a sensorimotor transformation, or a prospective output. The theoretical benefit of the affordance-competition view is that it makes such interactions testable rather than attributing all premotor activity to unspecified intention Cisek & Kalaska (2005) Graziano et al. (2002).
Preparation Without Overt Movement
Motor cortex is often active before electromyographic onset even when instructions require stillness. Preparatory firing therefore poses a mechanistic question: why does neural activity informative about a coming action fail to cause the muscles to execute it immediately? A population-level account proposes that preparatory states can evolve in dimensions that produce little or no drive on the output-potent dimensions linked to muscle activity. Kaufman and colleagues (2014) analyzed neural population activity in primates and presented evidence for such an output-null organization during movement preparation Kaufman et al. (2014).
Output-null and output-potent are relational concepts: the classification depends on a particular measured output and the mapping used to estimate it. A dimension null with respect to one muscle set might influence another population, a downstream circuit, or future activity. Furthermore, null-space analyses are usually built from limited neural samples and particular behavioral conditions, which leaves open the extent to which their geometries generalize across tasks and species. The important experimental result is a partial decoupling between robust preparatory activity and measurable movement, not proof that the preparatory network is causally sealed off from the body Kaufman et al. (2014) Shenoy et al. (2013).
Preparation can serve several functions simultaneously. It can reduce response latency after a go signal, constrain the trajectory of subsequent neural evolution, stabilize a preferred action, and coordinate the timing of descending control. Distinguishing these possibilities requires comparing changes in readiness with changes in execution under targeted intervention. Delaying a response does not necessarily impair the accuracy of the chosen movement; conversely, disturbing preparatory circuits may change reaction time and final trajectory together. The relation between preparation and execution is thus best assessed as a dynamical transformation between population states and behavioral output Churchland et al. (2012) Kaufman et al. (2014).
Motor Cortical Population Dynamics and Rotational Structure
Single-neuron discharge frequently varies in complex, context-dependent ways during reaching. Churchland and colleagues (2012) analyzed simultaneous and aggregated population recordings and found structured dynamical patterns in the movement-related activity of monkey motor cortex. Their analysis illustrated how population trajectories can display lawful rotational components even when limb kinematics are not themselves periodic. This finding encouraged a shift from asking which physical quantity each neuron exclusively represents to asking how a network's evolving state can generate appropriately timed output Churchland et al. (2012).
Dynamical systems models describe neural activity as a trajectory through a high-dimensional state space. A state may be shaped by recurrent connections, external inputs, proprioceptive feedback, and task-dependent context. Latent dimensions can summarize a population more effectively than analyzing every neuron in isolation, but dimensionality reduction is not a circuit mechanism by itself. The axes and rotations extracted by a model depend on sampling, preprocessing, trial alignment, and assumptions about noise. Competing dynamics can create similar low-dimensional trajectories; causal inference requires perturbation of the system and tests of recovery or deviation Shenoy et al. (2013) Gallego et al. (2017).
The explanatory promise of population dynamics is that it joins preparation, initiation, and execution within a single time-varying model. An action can be prepared by placing cortical activity in a favorable initial state, then unfolded through trajectories that recruit downstream muscle patterns. The same framework permits external drive to influence ongoing dynamics, so it does not require exclusively autonomous cortical oscillators. A complete biological model must eventually identify the afferent and recurrent pathways that shape state transitions. Recent perturbation experiments have made this distinction between internally sustained dynamics and input-driven activity experimentally significant Sauerbrei et al. (2020) Churchland et al. (2012).
External Drive, Thalamocortical Inputs, and Causal Dynamics
The causal architecture underlying motor cortical trajectories can be tested by transiently interrupting parts of the network during skilled movement. In a mouse forelimb task, Sauerbrei and colleagues (2020) used targeted perturbations to examine whether motor cortical patterns could recover through local dynamics alone or required ongoing input. Their results emphasized the importance of thalamocortical input for generating the activity associated with dexterous movement. This observation challenges simple autonomous-generator versions of the dynamical account while remaining compatible with structured population dynamics influenced by distributed inputs Sauerbrei et al. (2020).
Thalamus is a conduit for several functionally different pathways, including cerebellar and basal ganglia influences. Disrupting a thalamic input can alter cortical activity because it changes preparatory context, ongoing drive, recurrent stability, or access to task-related signals. The observed deficit alone does not determine which computation has been lost. Experiments become more discriminating when perturbation timing is varied and when effects are measured simultaneously in cortex, thalamus, and muscle. Loop-level hypotheses predict that specific interventions will change the evolution of population states, not merely their average firing level Gao et al. (2018) Bostan & Strick (2018).
Population models and pathway-level interventions should be regarded as complementary. Latent neural trajectories provide a compact description of coordinated temporal organization; anatomically restricted experiments identify the routes by which that organization can be sustained or interrupted. A coherent theory predicts both the geometry of neural change and the behavioral consequences of manipulating a circuit. It also specifies recovery mechanisms, compensatory activity, and the timescales over which an intervention acts. This approach offers a more informative test of motor control than assuming that high decoding accuracy settles the question of physiological implementation Shenoy et al. (2013) Miri et al. (2017).
Motor Cortex, Effectors, and Muscle Recruitment
Movement-related cortical activity is transformed by downstream circuitry into muscle recruitment. This transformation can be highly selective: the same broad neural area contributes differently to different behavioral contexts, and short-latency descending effects may depend on the prepared movement. Miri and colleagues (2017) studied the engagement of fast effector pathways and supplied evidence that motor cortex can recruit behavioral output through context-dependent mechanisms. The finding matters because it connects representations and population states to a measurable causal pathway rather than treating cortical activity as a complete description of movement Miri et al. (2017).
Reaching and grasping require distinct combinations of hand, forearm, shoulder, and axial muscles. A cortical population can contain information about a desired endpoint while downstream pathways manage posture-dependent joint torques and antagonistic coordination. The relation from neural signal to limb movement thus depends on spinal transformation and musculoskeletal mechanics. Long-train stimulation evoking complex postures is informative about access to coordination patterns, but electrical spread and stimulation duration prevent equating every evoked posture with a naturally selected action. Convergent recording and controlled movement tasks are needed to identify the function of the same pathway during ordinary performance Graziano et al. (2002) Loeb et al. (1999).
The most useful behavioral dissociations separate movement direction, movement timing, effort, and precision. A manipulation that changes peak velocity while preserving directional choice supports a different causal inference from one that alters which target is reached. Changes in reaction time can be caused by preparation, attentional access, response threshold, or downstream initiation delays. Simultaneous electromyography and kinematic tracking help distinguish these possibilities. Neural recordings add temporal detail but must be interpreted through physiological conduction and muscle activation delays. These methodological distinctions become central when cortical results are compared with manipulations of basal ganglia circuits Turner & Desmurget (2010) Thura & Cisek (2017).
Cortical Representations, Manifolds, and Learning Constraints
Neural population activity often occupies a limited region of the much larger mathematical space formed by recorded neurons. Gallego and colleagues (2017) described neural manifolds as a framework for understanding structured movement-related population activity across tasks. A low-dimensional manifold may reflect common input, recurrent connectivity, learned coordination, and restrictions imposed by the body. Its scientific usefulness rests on repeatable structure and predictive relationships, not on assigning a fixed semantic label to each latent axis. Distinct movements can recruit overlapping activity subspaces, and learning may reorganize how existing patterns are used Gallego et al. (2017).
Closed-loop interfaces permit investigators to modify the mapping from neural activity to external movement without changing a biological limb. Sadtler and colleagues (2014) studied how monkeys learned altered brain-computer interface mappings, reporting that changes compatible with existing patterns of neural covariation were learned more readily than changes requiring new patterns. The result suggests that network organization constrains rapid adaptation. It does not imply that difficult patterns are permanently inaccessible or that all forms of natural motor learning use identical constraints; the experimental task and training duration delimit the inference Sadtler et al. (2014).
Neural manifolds consequently provide a bridge between computational theory and rehabilitation technology. A decoder can exploit stable population structure while changes in neural activity can themselves reflect learning, attention, effort, or slow recording drift. The relevant scientific question is which patterns remain causally controllable across time and which changes reflect compensation rather than new motor skill. A neural interface that succeeds in communication demonstrates useful control of an external effector, while the conditions of performance differ from those of an intact limb under proprioceptive feedback. This distinction protects both the interpretation of experiments and claims about transfer Pandarinath et al. (2017) Sadtler et al. (2014).
Basal Ganglia Circuit Architecture and Historical Models
The basal ganglia contain interconnected structures including the striatum, pallidum, subthalamic nucleus, and substantia nigra. Their influence on movement involves loops with cortex and thalamus as well as projections to brainstem targets. Alexander, DeLong, and Strick (1986) presented an influential framework of multiple partly segregated circuits linking basal ganglia and cortex. This architecture helped explain why movement disorders can coexist with changes in cognition, motivation, and eye-movement control. The original account was a functional anatomical synthesis; direct observations of cell types, synaptic effects, and behavioral perturbations have since revised the simpler diagrammatic interpretation Alexander et al. (1986).
A major late-twentieth-century model distinguished a direct striatal pathway facilitating basal ganglia output disinhibition from an indirect pathway promoting suppression of inappropriate actions. Albin, Young, and Penney (1989) used this opposition to organize findings from movement disorders, and DeLong (1990) developed related circuit explanations from primate disease models. The model generated testable hypotheses about hypo- and hyperkinetic behavior and had lasting instructional value. Its limitations concern the oversimplification of pathway activity, movement context, and interactions with other circuits. A disease phenotype reflects chronic changes throughout a network and cannot be treated as a clean experimental lesion of one control computation Albin et al. (1989) DeLong (1990).
The physical arrangement of pallidal and nigral inhibition means that increased activity in one nucleus can facilitate a downstream target through disinhibition rather than direct excitation. This is an important mechanistic principle, but actual circuit effects vary with target population, receptor type, firing pattern, and behavioral timescale. Dopamine modulates striatal function across multiple receptors and pathways; its behavioral effects include changes in learning, vigor, and initiation. Therefore one neurotransmitter level cannot serve as a unique index of current action selection. Modern theories must reconcile anatomical circuit descriptions with recording and stimulation data collected during specific tasks Hikosaka et al. (2000) Klaus et al. (2019).
Direct and Indirect Pathways: Causal Experiments and Revision
Cell-type-targeted perturbation provided a stronger test of striatal pathway hypotheses than observations based exclusively on pathological lesions. In a parkinsonian mouse model, Kravitz and colleagues (2010) applied optogenetic control to direct and indirect basal ganglia pathways and observed distinct effects on motor behavior. The experiment gave causal support to pathway-dependent modulation under a defined disease-model condition. Yet selective stimulation is not equivalent to the natural spatiotemporal patterns normally present in a behaving animal, and artificial circuit activation may reveal available causal influence without identifying the full control code Kravitz et al. (2010).
Cui and colleagues (2013) recorded identified striatal pathway populations around self-initiated action and found concurrent activation of direct- and indirect-pathway neurons during movement initiation. This result challenges the literal interpretation that natural movement simply turns one pathway on while turning the other off. Coactivation can support selection of one behavioral channel while suppressing alternatives, regulate movement boundaries, or reflect distributed task-related computations. The observation alone does not choose among those possibilities, but it places an important constraint on any model claiming strict temporal opposition between the populations Cui et al. (2013).
The appropriate synthesis treats the classical direct/indirect account as a productive starting model whose causal details depend on behavior and recording scale. Pathway-level stimulation establishes that different routes can powerfully influence movement; natural population recordings demonstrate that these routes may be simultaneously engaged. To resolve their interaction, experiments must identify which action-specific ensembles fire, their timing relative to candidate commitments, and how local effects propagate to pallidal, thalamic, and cortical targets. A common error is to treat apparently contradictory studies as if they measured the same intervention and the same outcome when they often differ on both dimensions Kravitz et al. (2010) Cui et al. (2013).
What, Whether, and When: Initiation as an Independent Computation
Action initiation includes at least three questions: which action becomes available, whether a response is currently worthwhile, and when overt execution begins. Klaus, Alves da Silva, and Costa (2019) organized experimental evidence around these distinct problems. Their analysis makes timing and self-paced behavior central rather than viewing movement onset as the automatic result of a target choice. A striatal circuit can affect the probability or timing of movement without encoding detailed joint trajectories. The theoretical challenge is to identify how motivation, sensory context, urgency, and motor readiness jointly determine the transition into action Klaus et al. (2019).
Behavioral tasks differ substantially in what 'initiation' means. In a reaction-time experiment, a go cue releases a prepared response; in self-paced movement, the animal chooses the onset without an explicit external signal; in spontaneous locomotion, postural transitions and motivational state complicate temporal measurement. An experimental change in start latency therefore need not imply a change in subjective willingness or abstract intention. Analyses should consider latency distributions, movement amplitude, trials on which no movement occurs, and task-dependent preparatory activity. These measures help discriminate changes in motor availability from changes in decision rule or action vigor Dudman & Krakauer (2016) Turner & Desmurget (2010).
Anatomical pathways can also support decisions before movement begins by changing the availability of motor routes. The basal ganglia project to brainstem mechanisms capable of influencing eye and body movement, while thalamic loops affect cortical preparation. A precise temporal account requires simultaneous measurements at several points in the loop because a change in striatal activity may precede cortical transitions yet be causally mediated through recurrent feedback. Timing alone is insufficient to demonstrate that one region 'issued the command.' The strongest interpretation ties onset-related activity to targeted perturbations and the resulting change in action timing Hikosaka et al. (2000) Klaus et al. (2019).
Decision Commitment, Urgency, and the Speed–Accuracy Trade-off
Choosing between actions requires resolving uncertainty about alternatives, but the optimal moment of commitment can depend on time costs as well as evidence quality. The urgency concept captures increasing pressure to respond as deliberation continues. In primate reaching tasks, Thura and Cisek (2017) examined basal ganglia involvement and reported results supporting regulation of urgency rather than direct target choice in the studied circuits. Their finding challenges a general-purpose basal ganglia selector interpretation for these tasks and requires a sharper distinction between target preference and the threshold or timing of commitment Thura & Cisek (2017).
A crucial inference concerns what remains unchanged when neural activity is perturbed. If an intervention changes movement onset and the response distribution under time pressure while preserving information about the selected target in cortex, the result supports a role in commitment dynamics. But changes in motor execution itself can also affect measured reaction time and apparent speed–accuracy behavior. The experimenter must measure kinematics, error types, and evidence conditions rather than equating latency with deliberation. Models that fit response-time distributions should be evaluated against held-out conditions and independent recordings, since several parameter combinations can imitate a similar behavioral pattern Cisek & Kalaska (2005) Thura & Cisek (2017).
The urgency account is compatible with multiple interacting neural mechanisms. Cortical circuits may represent competing action possibilities while basal ganglia pathways modulate the likelihood or timing of the transition to action; thalamic and brainstem outputs then alter subsequent trajectories. This is a conditional synthesis, because basal ganglia subregions and species differ and some tasks require suppression of powerful prepotent responses rather than commitment among spatial targets. A robust theory must predict circumstances in which target coding, timing, and vigor dissociate, then identify the circuit-level reason for the dissociation Klaus et al. (2019) Dudman & Krakauer (2016).
Movement Vigor, Incentive Context, and Energetic Control
Motor vigor describes features such as speed, force, and willingness to expend effort within a task. It can change when the selected action remains the same. Dudman and Krakauer (2016) reviewed evidence that basal ganglia function may be understood partly through regulation of vigor rather than exclusive specification of movement content. Turner and Desmurget (2010) similarly emphasized that the effects of basal ganglia dysfunction on movement may arise through an impaired contribution to normal control and from secondary disruption in recipient structures. Both accounts caution against translating a generalized movement deficit directly into a single absent command Dudman & Krakauer (2016) Turner & Desmurget (2010).
Vigor is sensitive to reward, effort, fatigue, action costs, and behavioral context, yet its experimental measurement can be confounded by muscle weakness and motor impairment. A movement that is slower in disease may reflect changes in command gain, action selection, biomechanical control, or sensory correction. An experiment that varies incentives while holding movement requirements constant may test motivated vigor more cleanly than a comparison of unrelated tasks. If a circuit intervention modifies velocity and initiation independently, the two outcomes should be modeled as distinct parameters and then related to shared network dynamics Dudman & Krakauer (2016) Thura & Cisek (2017).
Neuromodulatory signals provide one route by which value and prior learning influence movement, but the precise mapping from dopamine activity to motor vigor remains an empirical problem. Dopamine's involvement in prediction and reinforcement learning cannot be used as a substitute for measurements of immediate performance. The strongest motor experiments track the same animal across variable motivational conditions, quantify both choice and movement, and compare intervention effects within defined pathway populations. The resulting framework connects action policy to execution intensity without collapsing neural control of current movement into the distinct problem of learning which rewards to pursue Klaus et al. (2019) Turner & Desmurget (2010).
Action Sequences, Saccades, and Effector-Specific Control
Motor behavior often unfolds in structured sequences rather than isolated movements. Jin, Tecuapetla, and Costa (2014) examined distinct basal ganglia subcircuits during learned action sequences and reported differential encoding of sequence boundaries and component organization. These findings support the idea that striatal circuitry can participate in parsing and concatenating movement elements. They do not demonstrate that every behavioral sequence is stored as a fixed motor program in the striatum; cortical, sensory, and spinal processes remain necessary for context-dependent execution and correction Jin et al. (2014).
Eye movements provide a particularly informative system because target choice and movement onset can be measured precisely. Hikosaka, Takikawa, and Kawagoe (2000) synthesized evidence that basal ganglia influence purposive saccades through projections affecting superior colliculus activity. This pathway illustrates how tonic inhibition and its release can regulate access to a downstream movement system. Eye-movement results cannot automatically be generalized to multijoint limb coordination, because effectors, brainstem circuitry, mechanics, and feedback constraints differ. Their comparative value lies in revealing testable control principles rather than establishing universal anatomy Hikosaka et al. (2000).
Sequence tasks and saccade tasks also reveal why action selection must be defined at the scale of the experiment. Selection may concern the next target, the start of a sequence, the choice of an entire behavioral routine, or the suppression of a competing response. Distinct neural mechanisms may dominate these levels, and changes in sequence fluency can reflect boundary timing rather than selection of a final goal. A useful comparative program combines the temporal precision of saccades, the rich kinematics of reaching, and the circuit specificity of self-paced animal movement to test whether shared principles survive differences in task structure Jin et al. (2014) Hikosaka et al. (2000).
Cerebellar Prediction, Timing, and the Internal-Model Hypothesis
Cerebellar circuits are implicated in precise movement coordination, temporal prediction, and adaptation when the relationship between intended commands and sensory consequences changes. The internal-model hypothesis proposes computations that estimate the future state or sensory consequences of an action and, in some formulations, map desired outcomes back into appropriate commands. Wolpert, Miall, and Kawato (1998) presented forward and inverse models as candidate explanations for movement under delayed feedback. These are computational hypotheses whose biological realization must be assessed through specific patterns of Purkinje-cell activity, circuit connectivity, and behavioral learning Wolpert et al. (1998).
A prediction of movement outcome can reduce the effect of sensory feedback latency, permitting control before delayed signals arrive. Yet the existence of useful predictions does not require a single centralized estimator. Cerebellar nuclei, cortex, thalamus, spinal circuits, and sensory pathways interact over several timescales, and multiple models can predict similar reaches. Ito (2008) examined an internal-model perspective on cerebellar function, including possibilities beyond overt movement. The broader theoretical proposal should be distinguished from the more directly supported role of cerebellar circuits in particular sensorimotor tasks, since general cognitive extrapolations depend on independent tests Ito (2008) Shadmehr et al. (2010).
A consensus review by Manto and colleagues (2012) documented the diversity of theories concerning cerebellar motor control, including timing, prediction, adaptation, coordination, and other computations. That diversity matters scientifically because accurate movement is an outcome shared by several possible mechanisms. To distinguish them, researchers can independently manipulate sensory delays, mechanical loads, perturbation predictability, or the timing of circuit disruption. An impairment in motor accuracy is informative about participation; a selective change in trial-by-trial predictions or error-dependent updating is more informative about the hypothesized computation. Such separation is fundamental to rigorous interpretation of cerebellar intervention studies Manto et al. (2012).
Purkinje Cells, Error Signals, and Movement Encoding
Purkinje cells form a major output channel of the cerebellar cortex and influence deep nuclei through inhibitory projections. Their complex and simple spike activity has been studied in relation to movement, prediction, and learning signals. Herzfeld, Kojima, Soetedjo, and Shadmehr (2015) examined Purkinje activity during rapid eye movements and investigated how population structure could encode movement-related information despite complex individual responses. The study provides a physiological anchor for discussing cerebellar action signals, while leaving open how results generalize from saccades to multijoint limb movement and from specific recorded zones to the entire cerebellum Herzfeld et al. (2015).
An experimental error can mean several different things. A sensory prediction error compares an expected observation with what actually occurred; a task error compares performance with a goal; a motor execution error concerns deviations in movement variables. These quantities often covary in ordinary tasks, so an adaptive change cannot by itself reveal which signal caused learning. Shadmehr, Smith, and Krakauer (2010) argued for analyzing error correction and predictive control together while recognizing their different computational roles. Useful experiments manipulate target error and sensory discrepancy separately, or change visual feedback while preserving movement success Shadmehr et al. (2010).
Causal claims about Purkinje cells require attention to circuit state. Suppressing a cell population can alter downstream nuclear activity, and changes in behavior may reflect acute disruption of control, long-term recalibration, or both. Recordings from a restricted zone cannot identify a universal code for movement; conclusions should report effector, task phase, sensory conditions, and the neural population sampled. Likewise, a decoder that extracts movement direction from a population is not proof that the cerebellum explicitly computes direction as an independent variable. The challenge is to identify which activity patterns are necessary for particular predictions and corrections Herzfeld et al. (2015) Manto et al. (2012).
Cortico-Cerebellar Loops and Preparatory Activity
Evidence increasingly places the cerebellum within circuits involved before movement execution. Gao and colleagues (2018) studied a cortico-cerebellar loop during motor planning and demonstrated that activity in this loop influences preparatory signals under their experimental conditions. Such findings broaden a traditional emphasis on online error correction and motor coordination. They also show why the timing of neural perturbation matters: an intervention during preparation may impair action readiness even when the same intervention at another task phase has different consequences Gao et al. (2018).
Cerebellar contributions to decisions can extend to trials that require temporal accumulation of sensory information. Deverett and colleagues (2018) examined a mouse evidence-accumulation task using recordings and perturbations of lateral cerebellar cortex and reported performance effects. The result supports a causal contribution in the studied task but cannot, on its own, establish a universal cerebellar decision algorithm. Motor performance demands, memory requirements, and altered timing are potential mediators; follow-up experiments must distinguish representational accumulation from motor preparation and expression of the eventual response Deverett et al. (2018).
Basal ganglia and cerebellar pathways are anatomically connected to broad cortical and thalamic systems, and they influence overlapping motor problems. Bostan and Strick (2018) synthesized evidence supporting an integrated network rather than wholly isolated loops. A central research opportunity is to ask whether one subsystem changes the predicted consequences of a movement while another changes its initiation threshold or gain, and whether these effects remain independent during simultaneous circuit perturbation. Answering this question will require circuit-specific recording, multi-site intervention, and behavior that separates movement speed, target selection, and corrective precision Bostan & Strick (2018) Gao et al. (2018).
Biomechanics, Muscle Synergies, and Spinal Coordination
A movement command reaches a physical plant with inertia, nonlinearity, tendon compliance, joint interaction torques, and sensory receptors distributed throughout the body. Much apparent complexity of observed movement originates from the mechanics of the limb, not solely from variability in cortical activity. Loeb, Brown, and Cheng (1999) developed a hierarchical perspective emphasizing muscles, spinal circuitry, sensory receptors, and descending inputs. Their framework explains why simplified cursor tasks can isolate central processes but cannot alone establish how a limb with multiple muscles maintains stability and exploits mechanical regularities Loeb et al. (1999).
Synergy theories propose that the nervous system recruits structured patterns of muscle activity rather than specifying each muscle independently on every occasion. Bizzi and colleagues (1991) emphasized biological computations emerging from spinal organization. Overduin and colleagues (2012) reported that microstimulation in a motor system can activate a restricted set of coordinated muscle patterns, offering a causal foothold for the synergy hypothesis. However, factor analysis may identify low-dimensional patterns even when no dedicated neural module exists. The scientific inference requires physiological and behavioral evidence that individual components are controllable, recombinable, and relevant during natural movement Bizzi et al. (1991) Overduin et al. (2012).
An action-selection model becomes incomplete when it ignores the mechanical feasibility of the chosen response. A target may admit many joint configurations and torque patterns; only some satisfy stability, effort, or injury constraints. Sensory feedback and learned task-dependent control can choose among those solutions dynamically. Accordingly, fixed motor templates are most defensible as restricted descriptions of repeated patterns rather than universal programs. Movement execution is a continual coordination problem involving the chosen goal, body state, spinal transformation, and environmental contact. This reasoning makes biomechanics part of the causal explanation of neural control Todorov & Jordan (2002) Loeb et al. (1999).
Optimal Feedback Control and the Structure of Correction
Optimal feedback control frames movement as the selection of control signals that reduce expected task-relevant costs while accounting for uncertainty and dynamics. Todorov and Jordan (2002) developed a theory of motor coordination in which feedback responses are structured by the task rather than correcting every deviation indiscriminately. The resulting 'minimum intervention' intuition explains why variability in dimensions irrelevant to success may be tolerated while errors threatening the goal are corrected. This is a powerful formal explanation of movement organization; it is not by itself a demonstration that the nervous system implements a particular optimal algorithm Todorov & Jordan (2002).
The theory yields observable predictions when tasks alter which movement variables matter. If an experimental target becomes wider along one axis, correction may become less strict along that axis while remaining precise along the constrained dimension. Control laws should predict such task-dependent selectivity more convincingly than a generic claim that the brain seeks optimality. Estimates depend on assumed cost functions, noise, dynamics, and control horizons; flexible model fitting can conceal identifiability problems. Mechanistic confidence improves when a model predicts a new perturbation condition without refitting and when neural evidence constrains plausible implementations Shadmehr et al. (2010) Todorov & Jordan (2002).
Fast feedback responses show that cortical networks can participate in corrections typically associated with short latency. Pruszynski and colleagues (2011) used neural recordings and perturbation paradigms to investigate integration across joints and implicated primary motor cortex in rapid multijoint responses. This finding helps bridge reflex-like timing and task-sensitive control. It also argues against a rigid separation in which the spinal cord handles all fast corrections and cortex controls only deliberate later responses. The exact pathway and latency boundaries still depend on the specific muscles, afferent signals, and mechanical conditions under study Pruszynski et al. (2011).
Motor Adaptation, Prediction Errors, and Recovery Across Trials
Movement control unfolds within a trial and changes across trials. Online feedback adjusts the current movement; adaptation modifies future control in response to consistent discrepancies. A subject learning to compensate for a force field or altered visual mapping can improve behavior through implicit recalibration, strategic adjustment, and altered control gains. Shadmehr, Smith, and Krakauer (2010) analyzed how prediction, feedback error, and adaptation interact across these timescales. The scientific boundary is important: repeated improvement in endpoint accuracy does not establish that a single neural plasticity rule explains the behavioral change Shadmehr et al. (2010).
Short-term adaptation and acquisition of complex motor skills are also distinct. An experimentally imposed visuomotor rotation can reveal how movement predictions are adjusted, while sport expertise involves perceptual judgment, action selection under pressure, coordination of many joints, and extensive experience. Similarly, neural population learning in a brain-computer interface isolates the ability to alter a particular decoded control mapping rather than the entire process of natural skill acquisition. Strong studies use retention, transfer, dual-task tests, and generalization across movement parameters to determine what has actually been learned Sadtler et al. (2014) Gallego et al. (2017).
An integrated account distinguishes what changes immediately after a perturbation from what changes after repeated experience. Altered motor output can arise through a new strategy, movement vigor, altered feedback responsiveness, or revised predictive estimates. The mechanisms should be separated through appropriately chosen behavioral manipulations and neural recording rather than inferred from net improvement alone. Cerebellar involvement is often a central hypothesis for prediction-based adjustment, whereas basal ganglia, cortical, and spinal pathways also affect the performance used to measure learning. Explaining their division of labor requires perturbation at multiple temporal scales Wolpert et al. (1998) Manto et al. (2012).
Movement-Specific Basal Ganglia Output Signals in Contemporary Experiments
The traditional disinhibition account holds that a pause in inhibitory basal ganglia output permits a selected downstream motor center to act. The interpretation becomes more demanding when a complex forelimb action consists of successive components. Falasconi, Kanodia, and Arber (2025) recorded from substantia nigra pars reticulata neurons in mice and combined those measurements with optogenetic interventions and downstream brainstem recordings. They observed movement-specific decreases and increases in output neuron firing, with precisely timed effects on distinct forelimb components. This evidence supports a granular bidirectional account in which inhibitory output circuits participate in both release and suppression of particular movements Falasconi et al. (2025).
The experiment advances the debate beyond a simple average firing-rate comparison between movement and rest. Specific output populations may pause for one movement and increase discharge during another, and a population can tile the course of behavior with heterogeneous changes. Such findings make cell identity, projection target, and movement phase central to functional interpretation. A claim that the basal ganglia provide a uniform global go signal would fail to predict the observed temporal and movement specificity. Nevertheless, optogenetic tests of mouse output pathways establish experimentally accessible causal roles under a defined reaching behavior, and their generalization to human voluntary acts remains a distinct question Falasconi et al. (2025) Hikosaka et al. (2000).
The new output-level evidence complements earlier direct/indirect striatal pathway experiments. Cui and colleagues documented concurrent recruitment of upstream pathways during initiation; Falasconi and colleagues directly linked downstream inhibitory output patterns with movement-specific motor control. These studies analyze different positions in the circuit, so their results are integrated by examining the transformation from striatal activity into temporally organized pallidal or nigral signals, then into brainstem and thalamocortical effects. A central unresolved task is to record these stages together during a behavior that separates chosen action, movement phase, initiation probability, and corrective effort Cui et al. (2013) Falasconi et al. (2025) Bostan & Strick (2018).
Discrete Action Counting and Continuous Goal-Directed Steering
A 2026 mouse study extended pathway-specific investigation into behavior that combines sequential action count with continuous steering. Fallon and colleagues trained mice to perform a specified number of lever presses for reward and measured both the discrete progression of the sequence and the continuous movement toward spatial goals. Pathway-selective optogenetic stimulation generated distinguishable changes: direct-pathway activation extended press sequences and biased steering in one direction, while indirect-pathway activation tended to terminate sequences prematurely and steer in the opposite direction. In the studied task, the two pathways exerted dissociable influences on multiple aspects of behavior Fallon et al. (2026).
The investigators also identified striatal populations with activity related to approach or count progress, and the difference between pathway population signals varied with proximity to goals. These findings provide a way to connect continuous movement coordinates with discrete behavioral progression. They refine the pathway debate because antagonistic causal effects under selected stimulation can coexist with concurrent natural recruitment and heterogeneous population responses. Rather than choosing between a pure antagonistic switch and a uniform coactivation model, experiments must specify the relevant state variable: sequence count, direction of steering, target commitment, or the timing of a motor component Fallon et al. (2026) Cui et al. (2013) Jin et al. (2014).
Interpretation depends on the experimental contingency linking movement to reward. A lever-counting task asks an animal to keep track of required repetitions, which is related to action control but not equivalent to freely choosing between independent goals in a natural environment. The effects of direct and indirect stimulation are causal for the task and parameters tested; their quantitative contributions to untrained behavior remain open. The key advance is a design that measures discrete and continuous components simultaneously, with sufficient temporal detail to ask how neural dynamics steer progress through a multistep behavioral objective Fallon et al. (2026) Klaus et al. (2019).
Learning-Dependent Thalamocortical Influence on Motor Execution
Motor experience alters not only local cortical representations but also the effective influence of upstream circuits. Ramot and colleagues (2025) used longitudinal axonal imaging, activity measurements, and optogenetic methods in mice learning a dexterous movement. They found that motor-thalamic inputs increasingly carried information about the learned movement and preferentially drove cortical neurons associated with expert performance. Inactivating those thalamic inputs impaired execution after learning. The result places experience-dependent, pathway-specific communication at the center of skilled motor output, rather than locating learning exclusively within motor cortex Ramot et al. (2025).
These observations clarify a significant ambiguity in population models. Stable cortical trajectories after training could arise from changes in local recurrent connectivity, changes in afferent drive, or both. Evidence that thalamic inputs gain selective influence and become necessary for expert execution strengthens the latter account in the studied mouse task. It fits with evidence for cortico-cerebellar loops supporting preparatory activity and for input-driven cortical pattern generation. Yet the studies vary in neural targets, training, and task phases, and their convergence should be framed as a distributed-circuit hypothesis rather than as proof that every learned skill depends on an identical thalamocortical mechanism Ramot et al. (2025) Gao et al. (2018) Sauerbrei et al. (2020).
Longitudinal design adds an evidential dimension unavailable in a cross-sectional comparison of experts and novices: a neural input can be examined before and after the same animal acquires a skill. Even so, the relation of circuit changes to retention, flexible transfer, and behavior under altered sensory feedback remains to be tested. The full experimental question is how distributed circuits acquire a reliable mapping from context and action policy to executable motor patterns. Combining learning-stage interventions with neural population analyses and detailed biomechanics can test whether a learned input changes selection, preparation, gain, or movement correction Ramot et al. (2025) Sadtler et al. (2014).
Human Incentive Context and the 2026 Movement-Initiation Evidence
Human movement initiation depends on motivational significance as well as current motor readiness. Dundon and colleagues (2026) combined high-field resting-state connectivity imaging with functional imaging during incentivized reaching to examine how ventral and dorsal putamen-centered circuits relate to affective and sensorimotor regions. Their results supported distinguishable patterns of engagement under reward- and loss-related conditions, accompanied by different effects on initiation and false starts. The study provides an important human systems-level test of how incentive valence may alter the circuits recruited during voluntary action Dundon et al. (2026).
The contrast between reward and avoidance conditions is especially informative for theories that equate one motor vigor signal with all motivation. Under the reported high-reward condition, movements began faster but false starts increased, while loss-related conditions displayed a different pattern of network involvement. These observations identify behaviorally relevant distinctions in urgency, inhibition, and incentive context. However, functional connectivity and task-related blood-oxygen-level signals remain indirect, correlational measurements of neural function; they do not establish that the proposed ventral-putamen route causally produces a particular motor decision. Selective pathway perturbations and replication in independent human samples would strengthen the mechanism Dundon et al. (2026) Thura & Cisek (2017).
Together, the 2025–2026 literature supports an increasingly differentiated account of movement control. Movement-specific output signals, pathway-specific effects on discrete and continuous actions, learned thalamocortical influence, and incentive-dependent human network engagement all challenge descriptions in which action selection and execution are governed by a single invariant gate. The studies also expose unresolved differences in explanatory scale: cellular circuit interventions in mice and hemodynamic network comparisons in humans constrain related but nonidentical hypotheses. The strongest next experiments will explicitly measure behavioral subcomponents, recording pathways, and cross-region influences under comparable action-selection tasks Falasconi et al. (2025) Fallon et al. (2026) Ramot et al. (2025) Dundon et al. (2026).
Experimental Paradigms, Measurement, and Causal Inference
Research on motor control gains explanatory power from combining tightly constrained tasks with measurements that preserve natural movement structure. Delayed reaches, instructed-delay saccades, self-paced locomotion, sequence tasks, force-field perturbations, and brain-computer interface control each isolate different parts of the problem. Delayed reaching is well suited to separating preparation from execution, while perturbation tasks expose correction and adaptation. However, the same numerical endpoint error may reflect different failures in target selection, muscle command, or online feedback. Task descriptions should therefore specify geometry, cue timing, movement onset criteria, feedback availability, and the instructed trade-off between speed and accuracy Cisek & Kalaska (2005) Pruszynski et al. (2011).
Neural recording provides rich observational evidence but can be affected by sample selection, common input, movement artifact, and the statistical consequences of averaging trials. Calcium signals have slower dynamics than spikes, and functional imaging indexes hemodynamic processes that are removed in time and scale from motor commands. Electromyography and kinematics offer direct behavioral constraints; force measurements reveal how similar trajectories can require different commands. Good analyses assess held-out prediction, report uncertainty, and compare plausible alternative mappings. A high-performance neural decoder cannot establish the necessity of the decoded signal for natural motor behavior without corresponding intervention Churchland et al. (2012) Gallego et al. (2017).
Causal interventions include lesions, pharmacological manipulation, electrical or magnetic stimulation, inactivation, and optogenetic perturbation. Their interpretability depends on targeting precision, timing, compensatory plasticity, and the physiological character of imposed activity. Optogenetic pathway effects are more cell-selective than many classical lesions, yet may activate patterns unlike those occurring naturally; stimulation can also reach axons traversing the target region. A rigorous causal analysis distinguishes necessity, sufficiency under artificial conditions, and mediation through a specific pathway. Combining targeted neural perturbation with trial-wise behavioral modeling yields stronger conclusions than attributing all observed impairments to the stimulated nucleus Kravitz et al. (2010) Sauerbrei et al. (2020).
Evidence Quality, Replication, and Cross-Species Inference
Motor neuroscience is methodologically heterogeneous. Important findings come from primates trained for precise reaching, rodents performing self-paced actions, human volunteers exposed to mechanical perturbations, and patients with movement impairments. Species differences affect corticospinal anatomy, limb use, behavioral repertoire, and the access researchers have to neural circuits. A direct/indirect pathway result in mice and a reach-target urgency result in primates address related functional questions but do not constitute simple independent replications. The proper evidence synthesis asks which causal propositions survive differences in species and task, and which remain restricted to an experimental preparation Cui et al. (2013) Thura & Cisek (2017).
Replication includes reproducing the behavioral contrast, the neural signal, the causal effect of an intervention, and the relationship among them. These are different levels of success. Independent reanalysis of population recordings can test robustness to preprocessing and dimensionality reduction; replication with different targets or movement speeds can test generalizability of a neural representation. Key statistical concerns include repeated measures, correlated trials, neuron-versus-animal unit of analysis, effect-size estimation, and the representativeness of highly trained subjects. Studies should also report exclusion criteria and failed or variable perturbations, which are often scientifically informative Shenoy et al. (2013) Sadtler et al. (2014).
Evidence grading in this narrative synthesis favors original manipulative studies for causal claims, well-controlled recordings for temporal and representational descriptions, and authoritative reviews for historical synthesis and conflicts across programs. It does not assign numerical certainty scores because such scores would require a formal prespecified evidence appraisal across a systematically identified corpus. In particular, a model that reproduces one task is treated as a useful explanation with defined scope, not as verified neural identity. Strong future work will make datasets, analysis code, and sufficiently detailed task parameters available to support independent tests of competing accounts Turner & Desmurget (2010) Manto et al. (2012).
Movement Disorders as Natural Experiments in Circuit Function
Disorders of movement provide compelling constraints on neurobiological theory because the affected circuits are necessary for everyday behavior. Parkinsonian bradykinesia, rigidity, altered initiation, and variability of movement highlight dysfunction in distributed basal ganglia-related networks. Classical pathway models helped organize disease manifestations and generated predictions about circuit balance. Nevertheless, chronic pathology modifies dopamine function, cortical state, thalamic transmission, compensatory strategies, and peripheral performance over time. A clinical symptom is therefore a joint network outcome rather than a selective intervention into one hypothetical motor decision variable Albin et al. (1989) DeLong (1990).
Interpretation also depends on what is being measured. Slowed movement may reflect reduced vigor, delayed commitment, inadequate preparatory scaling, or physical difficulty executing a selected command. Two people exhibiting similar reaction times can differ in acceleration profiles, corrective responses, and movement initiation thresholds. Similarly, cerebellar dysfunction can produce inaccurate timing and coordination with variable adaptation depending on the affected circuitry. Comparative clinical research must separate disease stage, medication state, task difficulty, fatigue, and coexisting sensory or cognitive deficits before attributing behavioral differences to a particular computation Dudman & Krakauer (2016) Manto et al. (2012).
Pathway manipulation experiments provide mechanistic insight but must not be mistaken for individualized treatment evidence. Optogenetic results in parkinsonian mouse preparations establish causal possibilities at the studied circuit and species level, not clinical efficacy of a particular therapy in humans. Human invasive and noninvasive interventions, where appropriate, require independent assessment of safety, comparison groups, durability, and functional outcomes. This dossier addresses the neuroscience of action, not personal diagnosis or a treatment protocol. The translational priority is to identify reproducible circuit-behavior relationships and test whether interventions restore meaningful behavior under everyday conditions Kravitz et al. (2010) Turner & Desmurget (2010).
Brain–Computer Interfaces and Artificial Action Systems
Brain-computer interfaces are experimentally valuable because they provide an alternative motor output channel with a measurable mapping between neural population states and external effects. Pandarinath and colleagues (2017) demonstrated high-performance communication by people with paralysis using an intracortical interface, illustrating that movement-related neural signals can support functional control when the usual muscular route is unavailable. This is direct evidence for useful neural decoding and closed-loop learning under defined interface conditions. It does not entail that the decoded dimensions are the full endogenous commands normally used for limb movement or that an interface restores all proprioceptive functions Pandarinath et al. (2017).
Closed-loop neural control also creates a controlled arena for testing how artificial mappings reshape learning. Sadtler and colleagues (2014) showed constraints on learning novel population patterns; Gallego and colleagues (2017) positioned such work within a broader framework of neural manifolds. Changes in decoding performance can arise from adaptation of the user, recalibration of the decoder, or both. Claims about retained neural identity need longitudinal recording stability and tests separating learning from algorithmic drift. The experimental benefit is the ability to manipulate output mappings that would be difficult or unethical to impose on ordinary biological movement Sadtler et al. (2014) Gallego et al. (2017).
In the Artificial Era, action-selection research also informs comparisons between biological and engineered agents. A robot controller can implement candidate evaluation, switching, trajectory planning, prediction, and error correction as formally distinguishable operations. Such engineering comparisons can clarify computational hypotheses about biological control when tasks, uncertainty, mechanical constraints, and interventions are matched. Functional similarity alone does not establish that the same physical mechanism or psychological property is present in both systems. The productive research question asks which control principles generalize across embodiments and which depend on vertebrate circuit architecture, development, and physiology Todorov & Jordan (2002) Loeb et al. (1999).
Major Scientific Disagreements and Unresolved Problems
A fundamental dispute concerns whether the basal ganglia select specific actions or regulate the conditions of commitment and vigor. Target-selection interpretations derive from circuit anatomy and movement-disorder evidence; urgency accounts emphasize experimental dissociations in primate reaching. These explanations are partly reconcilable because distinct basal ganglia channels, timescales, and effectors may serve different functions. The decisive experiment would manipulate identified pathway ensembles while recording cortical alternatives and downstream output in a task that separates target identity from urgency and velocity. The literature has established strong component results but has not yet delivered one universally validated cross-task algorithm Alexander et al. (1986) Thura & Cisek (2017).
A second dispute concerns what exactly population activity represents. Movement-direction tuning, dynamical state-space descriptions, and output-null preparation explain complementary observations, yet each can be overinterpreted when its measurement assumptions are hidden. The neural response of an individual cell may depend on posture and force while a low-dimensional model captures lawful temporal structure at the population level. An intervention that preserves one representation while changing the other could constrain their causal relationship. Likewise, evidence for input-driven cortical pattern generation requires models that incorporate recurrent dynamics and inter-regional drive rather than choosing between them by terminology Georgopoulos et al. (1982) Kaufman et al. (2014) Sauerbrei et al. (2020).
A third dispute concerns the function of predictive computations across cerebellum, cortex, and spinal networks. Internal models, optimal feedback control, and modular muscle synergies differ in level of description and may coexist, but each needs distinct falsifiable predictions. Determining whether a perturbation primarily changes prediction, feedback gain, or movement strategy requires coordinated neural, kinematic, and sensory measurements. Open questions include the developmental acquisition of action selection, circuit plasticity after injury, inter-individual variability, the consequences of naturalistic multitasking, and the generalizability of animal findings to daily human skills. Methodological progress will come from preregistered comparisons and shared multimodal datasets Wolpert et al. (1998) Todorov & Jordan (2002) Overduin et al. (2012).
Conclusion
Neural control of movement and action selection is a distributed biological process in which candidate actions, motor readiness, commitment, vigor, muscle recruitment, prediction, and error correction become coordinated across cortical, basal ganglia, cerebellar, thalamic, brainstem, and spinal systems. The experimental literature supports separable operations because target choice, response timing, movement intensity, and corrective precision can be dissociated. Anatomical connectivity and observed coactivation nevertheless show that these operations are implemented through interacting circuits. The strongest explanations identify both a behavioral computation and the conditions under which a defined neural manipulation changes that computation Klaus et al. (2019) Bostan & Strick (2018).
The history of motor neuroscience illustrates a cumulative change in explanatory scale. Spinal physiology established integration and inhibition; primate cortical recordings connected neural discharge to voluntary force and direction; basal ganglia circuit studies generated causal models of gating; cerebellar research supplied hypotheses about prediction and adaptation; and modern population analyses revealed structured preparatory and movement-related dynamics. Targeted perturbations have corrected simple interpretations of these findings, particularly the idea that direct and indirect striatal pathways alternate as universal on/off switches. Progress depends on treating each empirical result at its appropriate level of generality Eccles et al. (1956) Evarts (1968) Cui et al. (2013) Churchland et al. (2012).
The resulting scientific agenda centers on multi-region, multi-timescale causal experiments linked to explicit behavioral models. Future studies should compare candidate-action representations with commitment thresholds, resolve how cerebellar and basal ganglia influences converge on cortex, determine the conditions of rapid sensory correction, and quantify adaptation and recovery beyond restricted laboratory tasks. Robust movement science can also guide assistive interfaces and comparative artificial controllers through careful functional analogy and independent validation. This monograph establishes a verifiable synthesis of research questions, methods, and evidence while preserving the distinction between established neural effects and still-contested accounts of their computational organization Shadmehr et al. (2010) Pandarinath et al. (2017).
References
- Albin, R. L., Young, A. B., & Penney, J. B. (1989). The functional anatomy of basal ganglia disorders. Trends in Neurosciences, 12(10), 366–375. https://doi.org/10.1016/0166-2236(89)90074-X
- Alexander, G. E., DeLong, M. R., & Strick, P. L. (1986). Parallel organization of functionally segregated circuits linking basal ganglia and cortex. Annual Review of Neuroscience, 9, 357–381. https://doi.org/10.1146/annurev.ne.09.030186.002041
- Bizzi, E., Mussa-Ivaldi, F. A., & Giszter, S. (1991). Computations underlying the execution of movement: A biological perspective. Science, 253(5017), 287–291. https://doi.org/10.1126/science.1857964
- Bostan, A. C., & Strick, P. L. (2018). The basal ganglia and the cerebellum: Nodes in an integrated network. Nature Reviews Neuroscience, 19(6), 338–350. https://doi.org/10.1038/s41583-018-0002-7
- Churchland, M. M., Cunningham, J. P., Kaufman, M. T., Foster, J. D., Nuyujukian, P., Ryu, S. I., & Shenoy, K. V. (2012). Neural population dynamics during reaching. Nature, 487, 51–56. https://doi.org/10.1038/nature11129
- Cisek, P., & Kalaska, J. F. (2005). Neural correlates of reaching decisions in dorsal premotor cortex: Specification of multiple direction choices and final selection of action. Neuron, 45(5), 801–814. https://doi.org/10.1016/j.neuron.2005.01.027
- Cui, G., Jun, S. B., Jin, X., Pham, M. D., Vogel, S. S., Lovinger, D. M., & Costa, R. M. (2013). Concurrent activation of striatal direct and indirect pathways during action initiation. Nature, 494, 238–242. https://doi.org/10.1038/nature11846
- DeLong, M. R. (1990). Primate models of movement disorders of basal ganglia origin. Trends in Neurosciences, 13(7), 281–285. https://doi.org/10.1016/0166-2236(90)90110-V
- Deverett, B., Koay, S. A., Oostland, M., & Wang, S. S.-H. (2018). Cerebellar involvement in an evidence-accumulation decision-making task. eLife, 7, e36781. https://doi.org/10.7554/eLife.36781
- Dudman, J. T., & Krakauer, J. W. (2016). The basal ganglia: From motor commands to the control of vigor. Current Opinion in Neurobiology, 37, 158–166. https://doi.org/10.1016/j.conb.2016.02.005
- Dundon, N. M., Rizor, E. J., Stasiak, J. E., Wang, J., Li, T., Sabugo, K., Villanueva, C., Barandon, P., Babenko, V., Beverly-Aylwin, R., Stump, A., Santander, T., Bostan, A. C., Lapate, R. C., & Grafton, S. T. (2026). Incentive valence differentially engages open- and closed-loop basal ganglia circuits during movement initiation. Proceedings of the National Academy of Sciences, 123(19), e2537314123. https://doi.org/10.1073/pnas.2537314123
- Eccles, J. C., Fatt, P., & Landgren, S. (1956). Central pathway for direct inhibitory action of impulses in largest afferent nerve fibres to muscle. Journal of Neurophysiology, 19(1), 75–98. https://doi.org/10.1152/jn.1956.19.1.75
- Evarts, E. V. (1968). Relation of pyramidal tract activity to force exerted during voluntary movement. Journal of Neurophysiology, 31(1), 14–27. https://doi.org/10.1152/jn.1968.31.1.14
- Falasconi, A., Kanodia, H., & Arber, S. (2025). Dynamic basal ganglia output signals license and suppress forelimb movements. Nature, 644, 749–758. https://doi.org/10.1038/s41586-025-09066-z
- Fallon, I. P., Roshchina, M., Hong, F., Fernandez, S., Ruan, S., & Yin, H. H. (2026). Striatal pathways dissociably control action counting and goal-directed steering. Nature Neuroscience, 29(8), 1920–1930. https://doi.org/10.1038/s41593-026-02330-z
- Gallego, J. A., Perich, M. G., Miller, L. E., & Solla, S. A. (2017). Neural manifolds for the control of movement. Neuron, 94(5), 978–984. https://doi.org/10.1016/j.neuron.2017.05.025
- Gao, Z., Davis, C., Thomas, A. M., Economo, M. N., Abrego, A. M., Svoboda, K., De Zeeuw, C. I., & Li, N. (2018). A cortico-cerebellar loop for motor planning. Nature, 563, 113–116. https://doi.org/10.1038/s41586-018-0633-x
- Georgopoulos, A. P., Kalaska, J. F., Caminiti, R., & Massey, J. T. (1982). On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. Journal of Neuroscience, 2(11), 1527–1537. https://doi.org/10.1523/JNEUROSCI.02-11-01527.1982
- Graziano, M. S. A., Taylor, C. S. R., & Moore, T. (2002). Complex movements evoked by microstimulation of precentral cortex. Neuron, 34(5), 841–851. https://doi.org/10.1016/S0896-6273(02)00698-0
- Herzfeld, D. J., Kojima, Y., Soetedjo, R., & Shadmehr, R. (2015). Encoding of action by the Purkinje cells of the cerebellum. Nature, 526, 439–442. https://doi.org/10.1038/nature15693
- Hikosaka, O., Takikawa, Y., & Kawagoe, R. (2000). Role of the basal ganglia in the control of purposive saccadic eye movements. Physiological Reviews, 80(3), 953–978. https://doi.org/10.1152/physrev.2000.80.3.953
- Ito, M. (2008). Control of mental activities by internal models in the cerebellum. Nature Reviews Neuroscience, 9(4), 304–313. https://doi.org/10.1038/nrn2332
- Jin, X., Tecuapetla, F., & Costa, R. M. (2014). Basal ganglia subcircuits distinctively encode the parsing and concatenation of action sequences. Nature Neuroscience, 17(3), 423–430. https://doi.org/10.1038/nn.3632
- Kaufman, M. T., Churchland, M. M., Ryu, S. I., & Shenoy, K. V. (2014). Cortical activity in the null space: Permitting preparation without movement. Nature Neuroscience, 17(3), 440–448. https://doi.org/10.1038/nn.3643
- Klaus, A., Alves da Silva, J., & Costa, R. M. (2019). What, if, and when to move: Basal ganglia circuits and self-paced action initiation. Annual Review of Neuroscience, 42, 459–483. https://doi.org/10.1146/annurev-neuro-072116-031033
- Kravitz, A. V., Freeze, B. S., Parker, P. R. L., Kay, K., Thwin, M. T., Deisseroth, K., & Kreitzer, A. C. (2010). Regulation of parkinsonian motor behaviours by optogenetic control of basal ganglia circuitry. Nature, 466, 622–626. https://doi.org/10.1038/nature09159
- Loeb, G. E., Brown, I. E., & Cheng, E. J. (1999). A hierarchical foundation for models of sensorimotor control. Experimental Brain Research, 126(1), 1–18. https://doi.org/10.1007/s002210050712
- Manto, M., Bower, J. M., Conforto, A. B., Delgado-García, J. M., da Guarda, S. N. F., Gerwig, M., Habas, C., Hagura, N., Ivry, R. B., Mariën, P., Molinari, M., Naito, E., Nowak, D. A., Oulad Ben Taib, N., Pélisson, D., Tesche, C. D., Tilikete, C., & Timmann, D. (2012). Consensus paper: Roles of the cerebellum in motor control—The diversity of ideas on cerebellar involvement in movement. Cerebellum, 11(2), 457–487. https://doi.org/10.1007/s12311-011-0331-9
- Miri, A., Warriner, C. L., Seely, J. S., Elsayed, G. F., Cunningham, J. P., Churchland, M. M., & Jessell, T. M. (2017). Behaviorally selective engagement of short-latency effector pathways by motor cortex. Neuron, 95(3), 683–696.e11. https://doi.org/10.1016/j.neuron.2017.06.042
- Overduin, S. A., d’Avella, A., Carmena, J. M., & Bizzi, E. (2012). Microstimulation activates a handful of muscle synergies. Neuron, 76(6), 1071–1077. https://doi.org/10.1016/j.neuron.2012.10.018
- Pandarinath, C., Nuyujukian, P., Blabe, C. H., Sorice, B. L., Saab, J., Willett, F. R., Hochberg, L. R., Shenoy, K. V., & Henderson, J. M. (2017). High performance communication by people with paralysis using an intracortical brain-computer interface. eLife, 6, e18554. https://doi.org/10.7554/eLife.18554
- Pruszynski, J. A., Kurtzer, I., Nashed, J. Y., Omrani, M., Brouwer, B., & Scott, S. H. (2011). Primary motor cortex underlies multi-joint integration for fast feedback control. Nature, 478, 387–390. https://doi.org/10.1038/nature10436
- Ramot, A., Taschbach, F. H., Yang, Y. C., Hu, Y., Chen, Q., Morales, B. C., Wang, X. C., Wu, A., Tye, K. M., Benna, M. K., & Komiyama, T. (2025). Motor learning refines thalamic influence on motor cortex. Nature, 643, 725–734. https://doi.org/10.1038/s41586-025-08962-8
- Sadtler, P. T., Quick, K. M., Golub, M. D., Chase, S. M., Ryu, S. I., Tyler-Kabara, E. C., Yu, B. M., & Batista, A. P. (2014). Neural constraints on learning. Nature, 512, 423–426. https://doi.org/10.1038/nature13665
- Sauerbrei, B. A., Guo, J.-Z., Cohen, J. D., Mischiati, M., Guo, W., Kabra, M., Verma, N., Mensh, B., Branson, K., & Hantman, A. W. (2020). Cortical pattern generation during dexterous movement is input-driven. Nature, 577, 386–391. https://doi.org/10.1038/s41586-019-1869-9
- 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
- Shenoy, K. V., Sahani, M., & Churchland, M. M. (2013). Cortical control of arm movements: A dynamical systems perspective. Annual Review of Neuroscience, 36, 337–359. https://doi.org/10.1146/annurev-neuro-062111-150509
- Thura, D., & Cisek, P. (2017). The basal ganglia do not select reach targets but control the urgency of commitment. Neuron, 95(5), 1160–1170.e5. https://doi.org/10.1016/j.neuron.2017.07.039
- Todorov, E., & Jordan, M. I. (2002). Optimal feedback control as a theory of motor coordination. Nature Neuroscience, 5(11), 1226–1235. https://doi.org/10.1038/nn963
- Turner, R. S., & Desmurget, M. (2010). Basal ganglia contributions to motor control: A vigorous tutor. Current Opinion in Neurobiology, 20(6), 704–716. https://doi.org/10.1016/j.conb.2010.08.022
- Wolpert, D. M., Miall, R. C., & Kawato, M. (1998). Internal models in the cerebellum. Trends in Cognitive Sciences, 2(9), 338–347. https://doi.org/10.1016/S1364-6613(98)01221-2