Dynamic target tracking control of manipulator based on deep reinforcement learning.

In order to reduce the repeated work intensity of drilling personnel and the risks faced by high-risk environments during oilfield exploitation, an intelligent manipulator is introduced into the drilling platform, and a dynamic target tracking control method of the manipulator based on deep reinforcement learning is proposed in this paper. In this paper The digital twin of the manipulator is constructed in the virtual drilling simulation environment, and multi-agent parallel training is carried out using the machine learning agents(ML-Agents) framework and proximal policy optimization(PPO) algorithm. The ideal training algorithm model is deployed to the virtual manipulator on the virtual drilling platform, and the virtual and real manipulator can track the dynamic target synchronously through serial communication. The experimental results indicate that synchronous tracking of dynamic targets with virtual and real robotic arms is feasible and accurate, providing a new approach for the digital twin application in the oil and gas industry. [ABSTRACT FROM AUTHOR]