Abstract
How do people steer through a complex scene? We attempt to predict route selection from on-line visual control strategies for steering toward a goal and avoiding an obstacle. First, we modeled the behavioral dynamics of steering and obstacle avoidance. Participants were instructed to walk to a goal or around an obstacle, whose initial heading angle and distance were varied. Testing was done in a 40 × 40 ft virtual environment with a head-mounted display (60 deg H × 40 deg V), while head position was recorded with a hybrid sonic/inertial tracking system (50 ms latency). We analyzed the observed path, time series of heading angle, and phase portrait of turning rate as a function of heading angle. We find that goals behave as point attractors and obstacles as repellors in the phase plane, whose strength depends on distance (or time-to-contact). The change in turning rate can be modeled as a function of current turning rate, current heading angle, and the exponential of object distance (R2 * 0.97). Second, we used this model to predict the route taken around an obstacle to reach a goal, by composing terms for the goal and the obstacle. As initial goal distance gets closer, participants increasingly cut in front of the “repelling” obstacle, consistent with the increasing “attractiveness” of the goal. Third, by additively combining terms for goals and obstacles, we attempt to predict routes through a complex array of objects. Thus far, route selection appears to be consistent with an on-line dynamical control strategy, making explicit planning based on a 3D world model unnecessary.
NIH EY10923, NSF LIS IRI-9720327.