Combining Q-Learning and the Hungarian Algorithm for Multi-Robot Systems in Dynamic Environments

University of Science and Technology – The University of Danang Da Nang, Viet Nam This email address is being protected from spambots. You need JavaScript enabled to view it.
University of Science and Technology – The University of Danang Da Nang, Viet Nam This email address is being protected from spambots. You need JavaScript enabled to view it.
University of Science and Technology – The University of Danang Da Nang, Viet Nam This email address is being protected from spambots. You need JavaScript enabled to view it.
University of Science and Technology – The University of Danang Da Nang, Viet Nam This email address is being protected from spambots. You need JavaScript enabled to view it.
University of Science and Technology – The University of Danang Da Nang, Viet Nam This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

This paper presents a hierarchical reinforcement learning framework for multi-robot systems in dynamic warehouse environments. The proposed approach integrates Q-learning at two levels: motion control and task reassignment. At the motion level, Q-learning adaptively adjusts heuristic coefficients in the Artificial Potential Field (APF) model to improve trajectory smoothness and collision avoidance. At the task level, Q-learning evaluates whether to accept new assignments proposed by the Hungarian algorithm in scenarios with moving targets.

Simulation results show that the proposed hybrid method reduces travel distance, shortens task completion time, and decreases collision events compared to conventional APF and heuristic approaches. These results demonstrate that combining reinforcement learning with classical optimization improves adaptability and overall system performance in dynamic multi-robot environments.