A distributed closed-loop probabilistic robust prioritized motion planning algorithm

This paper addresses the problem of large scale multi-agent motion planning in the presence of various uncertainties and under limited communication bandwidth. Obtaining an optimal solution while simultaneously addressing all the issues is a difficult problem. Towards this, we develop a decentralized motion planner that combines probabilistic approaches including the rapidly-exploring random tree (RRT), chance constraint, overlapping coefficient and birthday paradox. First, we use RRT along with the chance constraint approach to identify robust paths. Second, we use the overlapping coefficient technique to detect conflicts between multiple agent paths and third, we use the birthday paradox to detect conflicts between agents in a large team under communication bandwidth constraints. Finally, a priority based approach is employed to resolve conflicts. These steps are carried out sequentially. To manage the level of uncertainty in the path planner, we use a closed-loop system to predict future distributions. Our paper extends the prediction approach to nonlinear Gaussian systems. We describe details of integrating these techniques to create a complete framework for determining computationally efficient paths for large scale multi-agent systems under uncertainty.