Skip to content
Conference

A hybrid reinforcement learning framework for multi-robot task allocation in dynamic environments

Oct 2026 · The International Conference on Optical Imaging and Deep Learning · Vol 14494, pp. 1449412 - 1449412-9 · 0 citations · 15 references
Engineering

Abstract

Computing systems for multi-robot coordination face significant challenges in dynamic and uncertain environments, especially where real-time task allocation and adaptability are critical. This study presents a hybrid reinforcement learning framework that combines deep Q-learning, swarm-based regularization, and distributed attention mechanisms to support efficient and scalable task assignment among heterogeneous robotic agents. To rigorously benchmark the performance, the proposed method uses a high-fidelity simulation environment with random tasks and dynamic obstacles. The experimental results show that the hybrid method improves the inference speed, task completion rate and path cost compared with the centralized and pure decentralized baselines. The system exhibits strong scalability as the number of agents increases, and effectively copes with non-stationary hazards and sudden changes in task assignment. Ablation studies show that group regularization improves spatial diversity and collision avoidance, while the attention module helps policy adjustment and adaptive information sharing. In the current work, the assumptions of reliable inter-agent communication and identical hardware capabilities are limitations. The research results lay a technical foundation for the further integration of edge computing platforms and multi-modal perception. This will provide a new solution for multirobot elastic autonomous operation in complex real-world scenarios.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.