Navigating annoying environments through evolution

Autonomous robotic systems are often tasked in the role of actively searching to find a target or set of targets which are to be either rescued, observed, or destroyed. In carrying out these missions, the vehicle must be capable of dealing with dynamic and possibly adversarial environments, which tend to foil or disrupt its intentions. As a step in this direction, the paper describes the application of a path planning technique rooted in simulated evolution to a number of scenarios of increasing complexity, which attempt to model various aspects of such an environment. The results presented illustrate the ability of this algorithmic approach to efficiently search simultaneously in space and time to deliver feasible, near-optimal solutions to problems involving varying terrain, dynamic obstacles, and moving targets. In doing so, we highlight the features of the evolution-based approach which make it particularly attractive for handling environments of arbitrary complexity.