HBicNet：Multi-Agent Reinforcement Learning Algorithm
Applying the classic multi-agent reinforcement learning algorithm MADDPG and BiCNet to the AGV task allocation scenario, introducing the Information Potential Field optimization reward mechanism to accelerate training convergence.
Model Architecture Diagram：
Description： In the dynamic task allocation scenario for multiple AGVs, our goal is for the AGVs to self-organize and continuously transport more goods. If we directly apply the multi-agent bidirectional coordination network BiCNet for task allocation, there are issues such as sparse rewards and non-stationary environments. To address this, we propose a hierarchical multi-agent reinforcement learning algorithm HBiCNet. By decomposing the complex task allocation problem into two sub-problems: target selection and path planning, AGVs can move more purposefully, which improves the efficiency of the multiple AGV system’s transportation.