Adaptation of Obstacle-avoidance in the Face of Emerging Environmental Dynamics

The avoidance of static obstacles is relatively easy to implement as an animat behaviour, even where the animat receives realistic`sensory' input from the environment (e.g., simulated, infra-red proximity signals). 1,2,3] The behaviour can also be learned by a variety of methods. However, natural implementations of static-obstacle avoidance may break down if the objects in the environment have dynamic properties. Any introduction of object motion directly aaects the environmental cues which anticipate impending collisions. The paper shows that this happens in a systematic way. The statistical order of the sensory collision cues increases roughly in proportion with the level of obstacle mobility. Thus, adaptation of obstacle-avoidance behaviours (in animats) involves making a transition from strategies based on low-order cues to strategies based on high-order cues.