Robot Collision Avoidance with a Guaranteed Safety Zone and Randomized Symmetry Breaking

Abstract

Collision avoidance of moving systems is a wellstudied
problem. The use of an Artificial Potential Field function
is a popular approach to compute in real time a path that avoids
collision between agents. It involves the minimization of a
weighted sum of an attractive force and a repulsive force.
Previous studies consider these weights to be fixed design
parameters, to be determined experimentally. In particular, these
parameters do not change during the run of the algorithm. Our
main result is based on the observation that by dynamically
changing these parameters one can obtain a guarantee on a
minimum safety distance between the agents. Specifically, if the
agents compute their path by minimizing the potential field with
properly chosen weights, there will always be a guaranteed safety
distance between each pair of agents. Our earlier studies show
promising experimental results and we extended the studies on
avoiding trajectory symmetry.Our simulation validates our
model and demonstrated its effectiveness for a group of noncooperative
agents moving in a small area.