Brett Fajen¹ (); ¹Rensselaer Polytechnic Institute

Most visually guided locomotor tasks allow for a range of solutions. When steering through a series of waypoints, for example, an actor might move directly toward the nearest waypoint or take a more indirect path that better anticipates upcoming turns. Individual movement trajectories differ not only in how effectively they satisfy task demands, but also in their stability and controllability given the actor’s movement capabilities (e.g., how quickly they can move, turn). Together, these factors define the affordance landscape; that is, the space of possible actions. By learning the affordance landscape, actors can adapt behavior to task demands while maintaining stable control within their limits. The aim of this study was to extend existing information-based strategies of visual control to model how an agent might learn the affordance landscape for a basic locomotor task: steering to a goal. The model agent’s translational and rotational accelerations are regulated by a simple error-nulling control strategy based on visual information about goal-heading angle, rotation rate, and speed. The control strategy has parameters that the agent can tune to alter trajectory shape and enact different solutions. The agent’s movements are also influenced by motor noise and subject to action limits that shape the stability and controllability of individual solutions. Importantly, the controllability of the current solution is specified by the time evolution of the control strategy output (acceleration), which converges to zero only when the agent is following a stable, controllable trajectory. This information allows the agent to gradually learn a function from currently available visual information and control-strategy parameters to the controllability of individual solutions; in effect, the affordance landscape. Model simulations capture key aspects of human steering behavior that simpler models cannot, including the ability to adapt to changes in action capabilities and motor noise and to coordinate steering and speed control.

Acknowledgements: NSF 2218220