Loading...

Controller Design Based on Adaptive Pattern Generators for Human Spine Movements Using Reinforcement Learning

Mahmoudi Filabadi, Mohammad | 2021

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58098 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Sadati, Nasser
  7. Abstract:
  8. In this thesis, a novel adaptive controller is designed for the musculoskeletal system of the human spine, using pattern generators and reinforcement, learning. The controller is similar to the biological motor control of humans. The proposed controller consists of two parts: 1) adaptive pattern generators 2) regulator. The conditional generative adversarial nets play the role of pattern generators. We have proposed two approaches, namely direct adaptive and indirect adaptive, to adapt conditional generative adversarial nets to the system uncertainties. A system identifier is used for the indirect adaptive approach. However, in the direct approach, we utilize reinforcement learning without any system identifier. Another part of the control structure is a regulator that reduces the regulation error. Moreover, to optimize the regulator performance, three methods for optimal gain tuning of the regulator using reinforcement learning have been introduced. The capabilities of the proposed systems are evaluated by several simulations in the presence of noisy measurements, disturbances, and perturbations. The results indicate that the proposed controller has excellent performance. Obstacle avoidance and optimal redundancy handling are the other advantages of the proposed control structures. The proposed control algorithms provide new insight on motion control of musculoskeletal robotic systems and human-like robots. Also, It is useful for different applications such as predicting muscles activation patterns in biomechanics
  9. Keywords:
  10. Conditional Generative Adversarial Networks ; Reinforcement Learning ; Musculoskeletal System ; Spine ; Pattern Generators ; Optimal Gain Tuning

 Digital Object List

 Bookmark

No TOC