Multi-Agent Path Planning in Complex Multi-Obstacle Environment: A Reinforcement Learning-Based Formation Containment Method
- 29 April 2026
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Automation Science and Engineering
- Vol. 23 (15455955) , 9880-9894
Abstract
This paper addresses the multi-agent path planning (MAPP) problem for heterogeneous systems where only a subset of agents, termed leaders, can perceive obstacles. A novel hierarchical framework is proposed, which decouples high-level path planning from low-level path execution. At the high level, a Multi-Agent Soft Actor-Critic (MASAC) algorithm trains the leaders to collectively navigate and dynamically shape an obstacle-free “safe mobile corridor”. At the low level, a distributed consensus-based controller guides the sensor-less followers to remain within this moving corridor, effectively tracking the path planned by the leaders. This architecture significantly enhances sample efficiency and scalability while naturally supporting low-cost, heterogeneous agents. The efficacy of the proposed method is validated through numerical simulations and experiments on a physical multi-UAV platform.Keywords
Funding Information
- National Key Research and Development Program of China (2023YFE0112500)
- National Natural Science Foundation of China (62303159)
- National Natural Science Foundation of China (62503153)
- Natural Science Foundation of Henan Province (252300421231)
- Henan Science and Technology Research and Development Plan Joint Fund (235200810088)
- Key Research and Development Project of Hainan Province (241111211200)
- Key Research and Development Project of Hainan Province (251111211100)
- Top Young Talent in Technological Innovation Program of Zhongyuan
This publication has 44 references indexed in Scilit:
- Flexible multi-UAV formation control via integrating deep reinforcement learning and affine transformationsAerospace Science and Technology, 2024
- Adaptive optimal formation control for unmanned surface vehicles with guaranteed performance using actor‐critic learning architectureInternational Journal of Robust and Nonlinear Control, 2023
- Multi-agent hierarchical reinforcement learning for energy managementApplied Energy, 2022
- Remaining useful life prediction of bearings by a new reinforced memory GRU networkAdvanced Engineering Informatics, 2022
- StockNet—GRU based stock index predictionExpert Systems with Applications, 2022
- Consensus in multi-agent systems: a reviewArtificial Intelligence Review, 2021
- Multi-agent deep reinforcement learning: a surveyArtificial Intelligence Review, 2021
- A survey on multi-agent deep reinforcement learning: from the perspective of challenges and applicationsArtificial Intelligence Review, 2020
- Cooperative Tracking a Moving Target Using Multiple Fixed-wing UAVsJournal of Intelligent & Robotic Systems, 2015
- Distributed finite-time attitude containment control for multiple rigid bodiesAutomatica, 2010