Collision Avoidance Based on Robust Lexicographic Task Assignment

Abstract

Traditional task assignment approaches for multiagent motion control do not take the possibility of collisions into account. This can lead to challenging requirements for the path planning and guidance of the individual agents. To address this issue, we derive an assignment method that not only minimises the largest distance between an agent and its assigned destination but also provides local constraints for guaranteed collision avoidance. To this end, we introduce a sequential bottleneck assignment problem and define a notion of robustness of an optimal assignment. Conditioned on a sufficient level of robustness in relation to the size of the agents, we construct local time-varying position bounds for every individual agent. These local constraints are a direct byproduct of the assignment procedure and only depend on the initial agent positions, the locations that are to be visited, and a timing parameter. We show that no agent that is assigned to move to one of the target destinations collides with any other agent if all agents satisfy their local position constraints. We demonstrate the applicability of the method in a numerical case study.