Safe multi-agent reinforcement learning framework for coordinated control in multi-robot systems

doi:10.1016/j.eswa.2025.130895

Safe multi-agent reinforcement learning framework for coordinated control in multi-robot systems

2025-12-18

0

PRE

AI

DOI：10.1016/j.eswa.2025.130895

Original

Original request for help

Save

Abstract

En 中文

Multi-robot deployment scenarios require individual and group controls for monitoring, managing, and functional modifications of the autonomous elements. Machine learning algorithms enable multiple robots to learn, adapt, and respond to control inputs efficiently. This article presents a Multi-Agent Reinforcement Learning Framework (MARLF) to ensure coordinated control of multi-robot systems. The proposed framework allocated agents for learning, control, and responding based on the operational environments. Each Agent is linked through repetitive control instructions that initiate the join and disjoin operations. The proposed framework ensures maximum agent support for multiple robots by grouping their operations under a single agent over similar time intervals. The operation time of the robots need not be the same as the control dissemination time for which reinforcement agents are provided. Learning helps identify flaws/errors in control dissemination and execution, as observed in previous intervals. To address such errors, reinforcement (new agents) is allocated specifically for the error-prone control instructions at the time of coordination. Therefore, the proposed framework includes multiple operational control functions to improve the efficiency of multi-robot systems.

Safe multi-agent reinforcement learning framework for coordinated control in multi-robot systems

Abstract

Journal

Organization