Risk allocation strategies for distributed chance-constrained task allocation

Abstract

This paper addresses the issue of allocating risk amongst agents in distributed chance-constrained planning algorithms. Building on previous research that extended chance-constrained planning to stochastic multi-agent multi-task missions, this paper presents a framework for risk allocation and proposes several strategies for distributing risk in homogeneous and heterogeneous teams. In particular, the contributions of this work include: proposing risk allocation strategies that exploit domain knowledge of agent score distributions to improve team performance, providing insights about what stochastic parameters affect the allocations and the overall mission score/performance, and providing results showing improved performance over previously published heuristic techniques in environments with given allowable risk thresholds.