Non-model robot multi-peg-in-hole assembly control method capable of using environment forecasting for optimization

Abstract

The invention provides a non-model robot multi-peg-in-hole assembly control method capable of using environment forecasting for optimization, and belongs to the robot assembly technology field. The non-model robot multi-peg-in-hole assembly control method capable of using the environment forecasting for the optimization uses a universal operating value function for forecasting environmental related knowledge, designs a fuzzy logic system according to human assembly experience, uses learned knowledge forecast as input, and outputs parameters used for optimizing a non-model control algorithm, and when the number of iterations of a deep reinforcement learning network reaches the upper limit and assembly meets requirements, the deep reinforcement learning network after completing training is output and used for outputting assembly action during the multi-peg-in-hole assembly process. The non-model robot multi-peg-in-hole assembly control method capable of using the environment forecasting for the optimization can achieve optimization for the existing non-model robot control algorithm, and shortens time needed in multi-peg-in-hole assembly of a robot.

Description

technical field [0001] The invention relates to a non-model robot multi-axis hole assembly control method optimized by environment prediction, which belongs to the technical field of robot assembly. Background technique [0002] As robots are widely used in industry, robot automation assembly technology has huge market application prospects. At present, deep reinforcement learning networks have been widely used to solve the application scenarios of complex control of actual robots. Based on deep reinforcement learning networks, robots There is no need to model and analyze the contact state in the assembly process, but to learn the assembly skills directly from the environment by trying like a human. [0003] Aiming at the use of non-model learning algorithms for multi-axis hole assembly tasks in the industry at present, the deterministic-based The strategy gradient search algorithm realizes the control of the assembly action of the robot, and uses the traditional fuzzy forc...

AI Technical Summary

Problems solved by technology

Although the non-model control algorithm has been developed rapidly and more and more applications in the actual industry, but its effect depends on the need for a large number of samples to try, real application scenarios can not be used for a large number of attempts, which is the limitation of this Major Challenges in the Application of Model-Free Control Algorithms in Real Environments

Images