Multi-agent Motion Planning for Dense and Dynamic Environments via Deep Reinforcement Learning

This paper introduces a hybrid algorithm of deep reinforcement learning (RL)\nand Force-based motion planning (FMP) to solve distributed motion planning\nproblem in dense and dynamic environments. Individually, RL and FMP algorithms\neach have their own limitations. FMP is not able to produce time-optimal paths\nand existing RL solutions are not able to produce collision-free paths in dense\nenvironments. Therefore, we first tried improving the performance of recent RL\napproaches by introducing a new reward function that not only eliminates the\nrequirement of a pre supervised learning (SL) step but also decreases the\nchance of collision in crowded environments. That improved things, but there\nwere still a lot of failure cases. So, we developed a hybrid approach to\nleverage the simpler FMP approach in stuck, simple and high-risk cases, and\ncontinue using RL for normal cases in which FMP can't produce optimal path.\nAlso, we extend GA3C-CADRL algorithm to 3D environment. Simulation results show\nthat the proposed algorithm outperforms both deep RL and FMP algorithms and\nproduces up to 50% more successful scenarios than deep RL and up to 75% less\nextra time to reach goal than FMP.\n