A soft actor-critic based method for two manipulators coordinated lifting and balancing

Robotic arms are widely used in industrial fields because of their flexibility and satisfactory scalability. With the development of robotics, the demand for multi-robotic arm collaborative control tasks in the industry is also increasing. Maintaining balance during the coordinated lifting of objects by two robotic arms is not only beneficial to the protection of objects but also reduces the load of each robotic arm to achieve safe and effective results. The task described in this article requires two robotic arms to coordinatedly lift a pot and maintain the ball in the center of the pot. For this multi-objective task, problem difficulty and solution time increase, and the design of the strategy becomes complex. Reinforcement learning has the advantage of learning trajectory planning strategies on its own. In this article, an offline control strategy based on soft actor-critic (SAC) is introduced. This strategy can successfully solve the coordinated balanced lifting problem of two manipulators in a simulation environment. The strategy achieves a high success rate.