Development of RL DDPG Agent for Planning Energy-Efficient Trajectory of Collaborative Robot

In the global scientific community, there are already approaches to optimizing robot trajectories based on classical mathematical methods. These methods often require significant computational resources, especially when solving high-dimensional problems in real time. In this regard, the use of approximate solutions, such as neural networks and reinforcement learning methods, is becoming a promising direction. In this paper, we train the Reinforcement Learning Deep Deterministic Policy Gradient method for creating an agent for optimizing robot trajectories based on the energy consumption criterion. The agent demonstrated the ability to plan trajectories with minimal energy consumption, with an error of no more than 1.3%. This proves the effectiveness of the trained method and its applicability in real production conditions.