Decentralized Multi-Robot Pursuit-Evasion
Development of a decentralized multi-robot pursuit-evasion framework using deep reinforcement learning for cooperative target interception under limited communication and sensing.
The framework employs Soft Actor-Critic (SAC) with centralized training and decentralized execution (CTDE), enabling each pursuer to make independent decisions using only local observations. The system is designed to scale to multiple robots operating in dynamic environments where communication constraints and collision avoidance must be considered.
The project includes a high-fidelity simulation environment featuring curriculum learning, domain randomization, realistic robot dynamics, moving evasive targets, and cooperative multi-agent behaviours. The developed framework forms the foundation for future deployment on real robotic platforms including autonomous ground vehicles and aerial robots.
This research is supported by the Central Queensland University Internal Research Grant Scheme and forms part of an ongoing research program on decentralized autonomous systems, multi-robot coordination, and intelligent decision-making.