Evolving Fuzzy Control of Manipulator in Open Changing Scenarios Based on Offline and Online Fusion Learning

High-performance and compliant manipulator control requires accurate dynamics models. However, the real manipulators are in open changing scenarios, and a controller with fixed structure and parameter often fails to control the manipulator well. In this article, a new evolving control framework that combines offline and online learning to improve the accuracy and compliance of manipulators in open changing scenarios is developed. The joint framework consists of offline learning based on a deep fuzzy neural network (DFNN) and online learning based on an evolving fuzzy system (EFS), where online learning uses both indirect learning and direct learning based on an expert knowledge base. To evaluate the effectiveness of the proposed method, tracking control simulations are performed on a PUMA560 manipulator as well as experiments on a real-world six-degree-of-freedom manipulator. The results show that the proposed method can effectively improve tracking control accuracy, significantly reduce feedback control torque, and improve adaptability to changing environments.