Modeling the journey as well as the destination: a control theory account of rotational navigation

Navigation requires estimating heading and transforming these estimates into actions. Prior models explain how self-motion and landmark cues are combined into heading estimates, but less is known about how these estimates are iteratively transformed into motor commands to reach a goal. Here, we hypothesized that navigation operates as a closed-loop process in which ongoing movement is updated by sensory prediction errors. To test this hypothesis, participants performed a goal-directed rotation task in virtual reality. On select trials, visual landmarks were shifted during movement, inducing a prediction error between the heading expected from self-motion estimates and the heading observed from the shifted landmarks. In parallel, we developed a closed-loop model of turning behavior that represents heading and angular velocity as jointly estimated states over time. This model accounts not only for final position—the destination—but also for the movement dynamics that produce it—the journey. The model predicts that landmark-induced visual prediction errors should produce rapid corrective changes in movement. Participant turning behavior qualitatively paralleled these model dynamics: acceleration changed after visual feedback, with larger landmark mismatches producing larger corrective responses. Together, these findings suggest that naturalistic movement depends on continuously transforming heading estimates into motor command through closed-loop control.