Deep Reinforcement Learning-Based Calibration Optimization for Robotic Manipulators Under Dynamic Uncertainties

This article is aimed at improving the calibration process of robotic manipulators operating in variable and uncertain environments. A combined method based on Deep Reinforcement Learning (DRL) and the Adaptive Neuro-Fuzzy Inference System (ANFIS) is presented. The main objective of this study is to make calibration faster, more accurate, of higher quality, and less dependent on manual adjustments. In the proposed system, the DRL agent learns how to correct calibration errors through interaction with the robot model in a simulated environment, while ANFIS helps to adapt control parameters and maintain system stability during motion. The method was tested on the UR5 manipulator and a 3-DOF robotic arm. The obtained results showed that the calibration time was reduced by approximately 45%, and positioning accuracy improved by about 60% compared to traditional methods. The robot can automatically adapt to various payloads and minor disturbances without human assistance. These findings prove that DRL-based calibration supported by ANFIS can make robotic manipulators more autonomous, stable, and efficient for real industrial applications.

Ҳали таржима қилинмаган