Biologically-inspired heuristics for human-like walking trajectories toward targets and around obstacles

摘要

We describe simple heuristics, based on perceptual variables, that produce human-like trajectories towards moving and stationary targets, and around moving and stationary obstacles. Interception of moving and stationary objects can be achieved through regulation of self-movement to maintain a target at a constant eccentricity, or by cancelling the change (drift) in the eccentricity of the target. We first show how a constant eccentricity strategy can be extended to home in on optimal paths and avoid obstacles. We then identify a simple visual speed ratio that signals a future collision, and the change in path needed for avoidance. The combination of heuristics based on eccentricity and the speed-ratio produces human-like behaviour. The heuristics can be used to animate avatars in virtual environments or to guide mobile robots. Combined with higher-level goal setting and way-finding behaviours, such navigation heuristics could provide the foundation for generative models of natural human locomotion.