Abstract
The steering dynamics model (Fajen & Warren, 2003, 2007) successfully simulates steering and obstacle avoidance during visually-guided locomotion. The instantaneous direction of travel (heading) is the resultant of attractive and repulsive forces acting on the agent, and locomotor paths emerge on-line without explicit planning. However, the model currently treats all obstacles as points. Gérin-Lajoie & Warren (2007) found that modeling a barrier as a set of point obstacles fails to capture human circumvention behavior. Here we propose four barrier models and test them against human data.
The candidate models represent barriers as: (1) a set of point obstacles whose repulsion is a weighted distribution along the barrier length; (2) a solid obstacle whose repulsion is based on its visual angle; (3) a point obstacle at the barrier's center plus two competing aim points at either end; (4) an aim point at one end, selected to minimize the total amount of turning on the future path.
Participants walk in an immersive virtual environment to a goal post (initial distance 10 m) and are instructed to take their preferred route around a barrier. To collect basic data on barrier avoidance in two experiments, we manipulate barrier length (2, 3, 4, 5 m), initial distance (4, 6, 8 m), lateral position (0.3, 0.6, 0.9 m), and orientation (95, 115, 135, 155, 175 deg). We fit each model to the human data and compare simulated and observed paths. The results provide a test of the adequacy of the four candidates and their underlying concepts. A successful model of extended barriers would allow us to generalize the steering dynamics model to locomotor behavior in everyday cluttered environments.
References
Fajen BR, Warren WH (2003) JEP:HPP. 29:343–62.
Fajen, BR, Warren, WH (2007) EBR, 180:303–319.
Gérin-Lajoie M, Warren WH (2007) 6th VSS Meeting, Sarasota, Florida.