An integrative computational approach for obstacle avoidance during action selection.

A computational model of obstacle avoidance in cluttered target-reaching tasks captures human-like action selection, including obstacle-driven trajectory changes and a transient tendency to initiate toward obstacles.

• Why it matters: Action selection in cluttered environments requires balancing goal pursuit with obstacle avoidance, but the brain mechanisms coordinating these competing demands remain unclear.
• What they did: The model extends stochastic optimal control by combining separate control policies for target approach and obstacle avoidance, weighted via a unified “relative desirability” measure integrating values from goals, obstacles, and actions.
• The result: In simulated target-reaching tasks based on prior human studies, the framework qualitatively reproduces how obstacle properties shape trajectories and the transient initiation toward obstacles before avoidance, enabling future work on more naturalistic action selection.