Graph Potential Field Neural Network for Massive Agents Group-wise Path Planning

Yueming Lyu, Xiaowei Zhou, Xingrui Yu, Ivor Tsang

Abstract: Multi-agent path planning is important in both multi-agent path finding and multi-agent reinforcement learning areas. However, continual group-wise multi-agent path planning that requires the agents to perform as a team to pursue high team scores instead of individually is less studied. To address this problem, we propose a novel graph potential field-based neural network (GPFNN), which models a valid potential field map for path planning. Our GPFNN unfolds the T-step iterative optimization of the potential field maps as a T-layer feedforward neural network. Thus, a deeper GPFNN leads to more precise potential field maps without the over-smoothing issue. A potential field map inherently provides a monotonic potential flow from any source node to the target nodes to construct the optimal path (w.r.t. the potential decay), equipping our GPFNN with an elegant planning ability. Moreover, we incorporate dynamically updated boundary conditions into our GPFNN to address group-wise multi-agent path planning that supports both static targets and dynamic moving targets. Empirically, experiments on three different-sized mazes (up to $1025 \times 1025$ sized mazes) with up to 1,000 agents demonstrate the planning ability of our GPFNN to handle both static and dynamic moving targets. Experiments on extensive graph node classification tasks on six graph datasets (up to millions of nodes) demonstrate the learning ability of our GPFNN.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

We revise the paper according to the AE's and reviewers' suggestions.

- (1) We highlight the difference between our focused problem and the TAPF in the second paragraph on page 4 of our revised paper. The TAPF setting requires the agent to stop once it reaches the target. In contrast, we focus on the continual group-wise planning problems in which the agents that achieve (reach) the target can continue to search for the remaining targets. Moreover, our method supports handling dynamic moving targets, while most TAPF methods cannot.

- (2) We further evaluate our GPFNN in the TAPF setting as the AE and reviewers suggested. More details can be found in the section E on page 20 of the Appendix. We compared GPFNN with the TAPF baselines (CBS-TA and ECBS-TA) on 54x54-sized maze and 250x250-sized maze. The experimental results show our GPFNN can achieve consistently higher success rates compared with the baselines in the TAPF setting.

Assigned Action Editor: ~Sungsoo_Ahn1

Submission Number: 3709