Multi-objective cooperative transportation for reconfigurable robot using isomorphic mapping multi-agent reinforcement learning

Abstract

In this paper, we propose an Isomorphic Mapping Reconfigurable Multi-Agent Reinforcement Learning (IM-RMARL) framework, which is suitable for decision-making scenarios in reconfigurable multi-agent reinforcement learning. This method holds promising applications in fields such as logistics transportation systems and disaster relief. Classical multi-agent frameworks typically assume that the number of agents is fixed and remains constant throughout the training process. However, in practical applications involving reconfigurable robots, the number of agents may vary over time or according to task requirements. Additionally, classical frameworks often assume easy access to abundant experience data for training and optimization. However, in reconfigurable robot clusters, this assumption may not hold true as not all combinations exist within a single episode. Our approach effectively addresses these challenges by integrating agent mapping mechanisms and similar type of intelligent agents’ experience sharing mechanisms, which aid in handling dynamic agent counts and limited experience data. Our experimental results demonstrate the effectiveness of the proposed framework, the Utilization Rate of Transport Capacity of the IM-RMARL group reaches 0.82, and the Task Completion Rate reaches 0.92.