Loading...

Dynamic Modeling of Lower Limb Exoskeleton Robot and the Human User with Walker

Fazeli, Sina | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58503 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Behzadipour, Saeed
  7. Abstract:
  8. One of the effective methods for restoring motor abilities in patients with spinal cord injury is the use of lower-limb exoskeleton robots. The design of these robots faces multiple challenges, including the selection of mechanical parameters, control algorithms, and the motion patterns of active joints. The common approach for determining these parameters has been trial and error, which is both costly and time-consuming. Therefore, the use of computational models and numerical simulations has emerged as a more efficient alternative. However, most of the models developed in previous studies have been based on inverse dynamics, meaning that the motion data of the user and the robot were given as inputs, and internal variables such as joint torques or robot–user interaction forces were computed. It is evident that solving the forward dynamics of such models requires the development of a representation of the central nervous system and the role of user learning in controlling and guiding the upper body. Only one selected study from the same research group has incorporated the role of user learning in interaction with the robot by presenting a complete model of the ReWalk robot and a user walking with crutches. In this research, the model developed in that selected study was extended to provide a platform for analyzing and determining the optimal design parameters of the Exoped exoskeleton robot (developed by the Iranian company Pedasis) and users who walk with a walker. The model was developed in the Simscape environment of MATLAB and is referred to as a forward dynamics simulation, consisting of three main components: the lower limb-robot model, the robot controller, and the user controller. The main strength of this simulation lies in the user control model, which represents the role of the central nervous system using the TD3 deep reinforcement learning algorithm. For simplification, the user’s upper limbs were excluded, and the control of the substitute forces and torques applied from the torso to the pelvis referred to as TP forces and torques was assigned to the user control model to achieve the desired motion goals defined through reward functions. Therefore, the input of the simulation is the motion pattern of the robot active joints, and its output is the TP forces and torques. To bring the simulation results closer to real conditions and to validate them, the actual motion of the Exoped robot with a paraplegic user and a walker was recorded. Through inverse kinematics and inverse dynamics computations, the mechanical parameters and motion pattern of the Exoped robot active joints were applied to the forward dynamic simulation. A comparison of the pelvis kinematics showed that the average differences between the simulation results and the real motion data ranged from 1.8% to 47%. Moreover, the correlation coefficients between the four pelvis kinematics variables in the simulation and the real data ranged from 0.7 to 0.9, indicating a strong correlation. The average values of some simulated TP forces and torques exceeded their corresponding real motion values by more than 100%, which can be attributed to simulation limitations and the omission of the upper limb inertia effects. Furthermore, the values of all reward functions—except for the one related to minimizing TP forces and torques were 3% to 99% higher in the simulation compared to the real motion, which is considered a positive outcome. Overall, this simulation successfully reproduced the kinematics of the robot and user with acceptable accuracy and can be used as a tool for determining the optimal design parameters of the Exoped exoskeleton robot to enhance the mobility of walker assisted users
  9. Keywords:
  10. Exoskeleton ; Deep Reinforcement Learning ; Inverse Dynamics ; Dynamic Simulation ; Lower Limb Exoskeleton ; Forward Dynamics Simulation ; walker

 Digital Object List

 Bookmark

No TOC