Loading...

Design, Implementation and Control of A Hexapod Robot Using Reinforcement Learning Approach

Shahriari, Mohammad Ali | 2013

1335 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 45072 (58)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Khayyat, Amir Ali Akbar
  7. Abstract:
  8. Hexapod Robots give us the ability to study walking robots without facing problems such as stability. It has a great deal of exibility in movement even if some legs become malfunctioned or face some diculties. Control of legged robots compared to other kind of locomotions is di- cult and requires fairly heavy on-line computations to be done in real time. Therefore, a machine-learning solution may be more convenient in such prob- lems. One method of machine-learning which is used for legged robots with outstanding potential is Reinforcement Learning (RL). RL is a promising ap- proach for achieving complex robots control in dynamic environments. RL techniques are interesting subjects in cognitive sciences and control theory. In control theory, designing a system that completely works with perfect perfor- mance is quite dicult. When unforeseen errors or disturbances are involved with the system it becomes an exhaustive procedure to use control theory. By designing a system which has the capability of learning how to accomplish a task on its own, there would not be any need for calculating and prediction iv of complicated control algorithms. The ability of learning is a core factor of cognition in cognitive sciences. Free gait analysis of walking robot is one such complex problem. It is to study the learning ability of the robot and see the learning abilities of a six-legged walking robot in a dynamic environment. A simple RL approach is used to develop walking gaits for hexapod. A six- legged (hexapod) robot will be implemented with Reinforcement learning al- gorithms. The capability of a robotic hexapod agent learning how to walk is explored using only the ability to move its legs and tell whether it is moving forward. Thus, the hexapod can be seen as an analog for a biological subject which lacks all the instincts excluding the basics which may be seen in infants without external support or parental gure.
    In this thesis an 18 degrees of freedom robot entitled as "SiWaReL" is de- signed and built. The kinematics analysis of the hexapod design is presented using a modular view approach. A dynamic model of the hexapod is modeled using objective oriented programming to execute the the developed iterative learning procedure in simulation environment which is used for free gait gen- eration. The results of all formulation and learning control are applied to the prototype, and then veried. This thesis studies learning ability of hexapod walking robot to walk using RL techniques and implements the developed approaches and formulations on a designed experimental six-legged walking robot prototype. The prototype is used for demonstration of the results
  9. Keywords:
  10. Reinforcement Learning ; Intelligent Control ; Hexapod Robot ; Robotics ; Gait Analysis

 Digital Object List

 Bookmark

...see more