Research on Learning from Demonstration System of Manipulator Based on the Improved Soft Actor-Critic Algorithm
Ze Qin Cui, Kui Li, Zenghao Chen, Peng Bao, Lang Kou, Lang Xie, Yue Tang, Danjie Zhu · 2022 IEEE International Conference on Robotics and Biomimetics (ROBIO) · 2022
With the promotion of artificial intelligence technology, service robots play an increasingly important role in human life. Compared with industrial robots, the audience of service robots is mostly non-professionals, so robots need to have stronger learning ability and can perform related tasks through human demonstration. This paper designs and develops a robot Learning from Demonstration (LFD) system based on the improved Soft Actor-Critic algorithm. Through continuous maximum entropy inverse reinforcement learning, the human demonstration data can be transformed into a robot executable task model by training. Finally, it is verified via simulation and experiments that the robot can effectively reproduce the human demonstration task through LFD system training.