Learning to Avoid Collisions

Members of a multi-robot team that operates within close quarters must avoid collisions. The typical, simple collision avoidance method has the robot compute its distance to other robots and stop, even move away, when this distance falls below a fixed threshold. While this approach may skirt dis-aster, it may also reduce the team’s efficiency, when robots halt at length for others to pass by, or travel further to move around one another. This paper reports on experiments where a human trainer, through a graphical user interface, watches robots perform an exploration task. The trainer can direct the robots to halt before they collide, and then to resume once their paths are clear. Experiment logs record the robots’ states, and a classifier is learned to identify the states in which “halt ” and “resume ” commands are issued. Preliminary re-sults indicate that it is possible to learn a classifier that mod-els trainers well, and that different trainers consider different factors when making their decisions