Reinforcement Learning-Based Parallel Approach Control of Micro-Assembly Manipulators

Micro-devices are usually assembled by micro-assembly robot operating multi-manipulators in a narrow assembly space. To ensure assembly accuracy, manipulators are required to assemble multiple parts in parallel. However, in the traditional assembly, in order to prevent the parts from interfering, the movement trajectory of each manipulator must be manually input, which leads to low planning efficiency. In this paper, a multi-body spatial approach algorithm is established based on reinforcement learning methods, and a multi-body collision avoidance control method based on grid method and reinforcement learning is proposed, which realizes the purpose of efficiently generating the running trajectory and improving the planning efficiency on the premise that multi-parts achieve the target pose without interference. In addition, the calibration method of the simulation space coordinate systems and the Cartesian space coordinate systems is proposed, the motion trajectory in simulation space is transformed into the Cartesian space motion trajectory to control manipulators movement. Experimental results verify the effectiveness of the proposed method, and realize intelligent and safe parallel approaching of multi-manipulators.