Experiment of Cooperative Transportation using Multi-Robots by Multi-agent Deep Deterministic Policy Gradient.

Cooperative transportations using many robots are able to support various tasks as compared with a transportation using a robot. However, cooperative transportations have been used infrequently in industrial applications due to the complexity of a formation change considering the avoidance of other robots and environments. In this paper, a control system to search robots’ paths for a cooperative transportation using a multi-agent deep deterministic policy gradient (MADDPG) is proposed. MADDPG is a deep reinforcement learning method specialized for a multi-agent system to determine the effective path for making the formation. The feature of the system, each robot finds a trajectory and moves autonomously for a cooperative transportation. The effectiveness of MADDPG is evaluated by an experiment result of a formation change for a cooperative transportation using multi-robots.