DMP-Based Motion Generation for a Walking Exoskeleton Robot Using Reinforcement Learning

• Published in: IEEE transactions on industrial electronics (1982) Vol. 67; no. 5; pp. 3830 - 3839
• Main Authors: Yuan, Yuxia, Li, Zhijun, Zhao, Ting, Gan, Di
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial legs; Dynamic movement primitives (DMP); exoskeleton robot; Exoskeletons; Hip; Human motion; Joints (anatomy); Knee; Learning; Legged locomotion; Locomotion; locomotion trajectory; Pendulums; reinforcement learning (RL); Robot dynamics; Robot sensing systems; Robots; Rolling motion; Trajectories; Trajectory; Uncertainty; Walking; Yaw
• DOI: 10.1109/TIE.2019.2916396
• ISSN: 0278-0046

For the purpose of the assistance for human walking, this paper describes a novel coupled movement sequences planning and motion adaption based on dynamic movement primitives (DMPs) for a walking exoskeleton robot. The developed exoskeleton robot has eight degrees of freedom (DOFs). The hip and knee of each artificial leg can provide two electric-powered DOFs to flexion or extension, two passive-installed DOFs of the ankle are to achieve the motion of inversion/eversion and plantarflexion/dorsiflexion, and two passive DOFs of the hip are to achieve the motion of roll or yaw. A novel trajectory-learning scheme based on reinforcement learning (RL) combined with DMPs is presented for a lower limb exoskeleton robot, aiming to give assistance to human walking. In the proposed strategy, a two-level planning is designed. In the first level, the inverted pendulum approximation under the consideration of the locomotion parameters is utilized to guarantee the zero-moment point within the ankle joint of the support leg in the phase of single support. In the second level, the joint trajectories are modeled and learned by DMPs. Meanwhile, the RL is adopted to learn the trajectories for eliminating the effects of uncertainties in joint space. The experiment involving four subjects based on a lower limb exoskeleton robot demonstrates that the proposed scheme can effectively suppress the disturbances and uncertainties.