A new study suggests that the brain may not choose between goals in continuous, real-world behaviour as if it were selecting a single option from a menu. Instead, it may constantly blend several control strategies, adjusting their influence as circumstances change. In research published in Nature, Chericoni, Fine, Ismail and colleagues examined how humans pursue moving targets in a continuous prey-pursuit task. Their findings point to a neural architecture in which different brain regions divide the work of estimating the current situation, evaluating its value and switching between competing behavioural policies. The result offers a control-theoretic explanation for how people make flexible decisions while moving through dynamic environments.
Traditional theories of decision-making often focus on discrete choices: selecting one item, accepting or rejecting an offer or pressing one button rather than another. Those models have been extraordinarily useful, particularly in economics and laboratory neuroscience, where choices can be isolated and measured. Natural behaviour, however, rarely unfolds in such clean steps. A person chasing a moving object, navigating a crowded street or trying to complete a task while responding to changing opportunities must continuously alter speed, direction, attention and effort. The goal itself may shift during the behaviour, meaning that decision-making is not simply a sequence of separate choices. It is an ongoing process of controlling action under uncertainty.
The researchers approached this problem using ideas from control theory, a mathematical framework developed to describe how systems regulate themselves over time. In engineering, a controller compares the desired state of a system with its current state and calculates actions that reduce the difference. A spacecraft adjusts its trajectory, for example, while a thermostat regulates temperature. In biological behaviour, a controller could specify how an individual should move or act to pursue a particular objective. One policy might prioritize intercepting prey quickly, another might conserve energy, while a third might maintain a safe distance. Rather than relying on a single fixed policy, the brain could combine these controllers and continuously change their relative weights.
This proposed mechanism is described as compositional control. The central idea is that behaviour can be decomposed into a mixture of lower-level control policies, each associated with a distinct pursuit goal. A higher-level meta-controller determines how strongly each policy contributes at a given moment. Technically, the resulting action can be understood as a weighted combination of policy outputs: when the environment changes, the weights change, causing the animal or person to alter behaviour without having to construct an entirely new strategy from scratch. This provides a potentially efficient solution to complex tasks. The brain can reuse learned controllers while a supervisory system decides which combination is most appropriate.
The prey-pursuit task allowed the researchers to observe this process in a setting that was more continuous and dynamic than conventional reaction-time experiments. Participants had to track or pursue changing targets, requiring them to plan ahead while responding to new information. Their movements were analysed with a control-theoretic decomposition designed to identify the component policies underlying each pursuit strategy. According to the study, the behaviour was best explained by a meta-controller that directed a mixture of goal-specific controllers. In other words, participants did not simply alternate between rigid modes of behaviour. They appeared to adjust the composition of their strategy as the demands of the task evolved.
The neural findings revealed a division of labour across several brain systems. Activity in the anterior cingulate cortex, a region frequently associated with monitoring performance, conflict and the need for behavioural adjustment, predicted major changes in the blend of policies. This pattern is consistent with the anterior cingulate acting as a meta-controller. Rather than directly encoding every movement, it may signal when the current policy mixture is no longer adequate and initiate a substantial reweighting of control strategies. Such a mechanism could help explain how people rapidly reorganize behaviour when a target changes direction, a previously effective plan fails or the relative importance of competing goals shifts.
The hippocampus appeared to perform a different computational role. Hippocampal neurons encoded and updated a latent policy state that supported early planning, suggesting that this region may help estimate the hidden structure of the task before action unfolds. A latent state is an internal representation of variables that cannot be observed directly but must be inferred from experience, such as where the task is heading, which strategy is currently active or what future conditions are likely. In this framework, the hippocampus functions as a state-estimating controller: it maintains an internal model of the situation and updates that model as new evidence arrives. This could allow the brain to prepare an appropriate combination of policies before a visible behavioural switch occurs.
The orbitofrontal cortex showed yet another pattern. Although this region is often linked to flexible decision-making and changes in strategy, the findings were more consistent with it representing the current value structure of the task rather than directly switching policies. Value structure refers to the expected consequences of available outcomes, including their desirability, cost and relevance under present conditions. If the value of speed rises, for example, a fast-interception policy may become more influential; if energy expenditure or risk becomes more important, a more conservative controller may gain weight. The orbitofrontal cortex may therefore provide the contextual signal that tells other systems what matters now, without itself serving as the primary command centre for policy selection.
Together, the results suggest a tripartite organization for continuous goal-directed behaviour. The hippocampus estimates the hidden state and supports early planning, the anterior cingulate cortex monitors the need for major policy changes and coordinates the reweighting of control strategies, while the orbitofrontal cortex represents the current value landscape in which those strategies operate. This division does not imply that each region works in isolation. Real behaviour depends on communication among them and with motor, sensory and motivational systems. Nevertheless, the framework offers a clear computational map for understanding how the brain can remain both stable and flexible: reusable controllers provide consistency, state estimation supplies prediction, value signals define priorities and the meta-controller decides when the blend must change.
The study could have wide implications for neuroscience, artificial intelligence and clinical research. Many current models of decision-making are optimized for isolated choices, while robots and autonomous systems must operate continuously in uncertain environments. A compositional control architecture could allow machines to combine specialized skills and adjust them according to changing goals instead of relying on a single monolithic controller. In medicine, the framework may also help clarify why disorders affecting motivation, planning or cognitive flexibility can produce apparently disorganized behaviour. If value representation, state estimation or policy reweighting becomes disrupted, an individual might persist with an ineffective strategy, switch too readily or struggle to coordinate multiple goals. By treating behaviour as continuous control rather than a chain of disconnected decisions, the research offers a new way to study the algorithms behind everyday intelligence—and a striking glimpse of how the brain turns shifting priorities into fluid action.
Subject of Research: The neural and computational basis of continuous, goal-directed behaviour and compositional control during prey pursuit.
Article Title: Neural basis of compositional control
Article References: Chericoni, A., Fine, J.M., Ismail, T.S. et al. “Neural basis of compositional control.” Nature (2026). https://doi.org/10.1038/s41586-026-10896-8
Image Credits: AI Generated
DOI: https://doi.org/10.1038/s41586-026-10896-8
Keywords: compositional control, continuous decision-making, control theory, goal-directed behaviour, prey pursuit, anterior cingulate cortex, hippocampus, orbitofrontal cortex, neural policy, computational neuroscience
