Design and Huaman-Robot Interaction Control of Multi-Mode Lower Extremity Rehabilitation Robot


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date4 Jun 2021


Lower extremity paralysis has become increasingly common in recent years, but rehabilitation treatment costs much time and money due to the expensive and limited therapist labor. Consequently, rehabilitation robots have been developed to assist physiotherapy to overcome this problem. In most of current rehabilitation robots, the actuation lacks flexible controllability, and the training mode is monotonous. In this research, a multi-mode lower extremity rehabilitation robot is designed by equipping magnetorheological (MR) actuators that can generate highly instantaneous flexible motion with less power consumption. Besides, the human’s effect on the robot plays an important role. Several human-robot interaction control schemes are proposed using this robot to improve rehabilitation effectiveness. On this basis, this thesis is presented from different perspectives.

Firstly, the lower extremity rehabilitation robot with MR actuators is designed and implemented. The robot is designed in mechanical structure, such that it exhibits lower extremity exercise for patients in sitting/lying posture, with four degrees of freedom for each limb, including hip abduction/adduction, hip flexion/extension, knee flexion/extension and ankle dorsiflexion/plantarflexion. Particularly, the MR actuators are optimized to realize high torque output in compact size by using the finite element method. In addition, the characteristic of the MR actuator is analyzed theoretically and experimentally. By taking advantage of the MR actuators, the robot can adopt multiple working modes, including robot-active mode and human-active mode. In the robot-active mode, the MR actuator works as a clutch to transfer the motion to the robotic joint, and the robot leads the human leg to move. In the human-active mode, the human limb dominantly guides the movement of the robotic exoskeleton, and the MR actuator functions as a brake to help the user conduct anti-resistance training for muscle strengthening. After manufacturing the robot prototype, a biomechanical analysis is undertaken with the established human-robot coupling model, and the control system is completed for further exploration.

Secondly, a multi-modal human-robot interaction control based on human motion intention estimation is proposed. The human motion intention can be reflected by skin surface electromyography (EMG) signals. An EMG-driven musculoskeletal model is built to estimate the human joint torque, where a biomechanical analysis platform called AnyBody Modeling System is applied to optimize the model parameters. Then, an adaptive impedance control is proposed for selecting the exact working mode that fits the requirement of patients in which the automatic switch between modes is allowed. Therefore, the proposed controller can generate appropriate robotic assistance or resistance intensity according to the patients’ motion intention, thus realizing assist-as-needed training.

Thirdly, a robotic mirror therapy framework with reinforcement learning is established for hemiparesis rehabilitation. The mirror therapy is implemented with a bilateral master-slave robotic system. The master robot interacts with the functional limb (FL) and the slave robot is wearable for the impaired limb (IL). During therapy, the FL actively drives the master robot to move, and the IL mimics its action with the assistance of the slave robot. Most especially, the reinforcement learning controller is adopted to explore the optimal rehabilitation strategy for the IL. The learning controller aims to maximize the IL’s muscle activation and minimize the motion trajectory tracking error through mirror transmission. Additionally, an asynchronous deep deterministic policy gradient algorithm is utilized by redesigning the experience replay to accommodate different patients’ exercise on multiple robotic platforms in parallel. This scheme significantly speeds up the data collection and reduces the learning time. The experiments indicate that the proposed robotic mirror therapy system can provide voluntary training for different hemiplegic patients, and the optimal rehabilitation efficacy can be efficiently achieved whilst guaranteeing safety.

In summary, the proposed robotic system offers flexible training modalities and appropriate exercise intensity with adapted to human movement capabilities, thus improving the rehabilitation of patients with varying lower limb paralysis.

    Research areas

  • rehabilitation robot, magnetorheological actuator, human-robot interaction control, mirror therapy, reinforcement learning