Loading...

Design of a Two-Layer Dynamic Based-Trajectory Based Controller on a Pacing Quadruped Robot

Tahvilian, Ali | 2019

631 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 52395 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Alasti, Arya; Salarieh, Hassan
  7. Abstract:
  8. The accuracy, speed, and ability to do work in hazardous environments has led to the development of robots in everyday life. Accordingly, research on quadruped robots for use in military environments is expanding. Designing control algorithms for walking and stepping on quadruped robots is one of the most important parts of the design of these robots, which creates complex movements despite the lack of sophisticated equipment in the robots. One of the robot gaiting control methods is dynamic based method that is based on the robot’s dynamics and resistant to perturbations. Trajectory based method is another robot gaiting control method that is inspired from CPG algorithm and has learning ability without requiring sophisticated robot dynamics.In this project, an attempt has been made to design a two-layer controller for a walking quadruped robot utilizing the benefits of both dynamic based and trajectory based control methods. The upper layer of this controller is dynamic based which is designed on a quadruped robot for first time based on Grizzle method and its lower layer controller is a trajectory based controller with learning ability. The gating process is initiated by the upper layer in this control algorithm, and in the continuation of the motion, the lower layer control the gaiting process by learning signals that has been produced by upper layer controller until creation of perturbations. This reduces the computational burden of the controller processor. In the existence of disturbances, the control process is shifted to the top layer to compensate for disturbances and stabilize movement. By extracting the robot dynamics and designing the top layer controller, the quadruped robot was able to move stable at 2.6 m/s. A low-layer controller consisting of three parts of learning, critic agent and distributed controllers was designed that controller part could not produce stable motion in the robot, which could be the because of the impact phase and the lack of subsystem synchronous between distributed controllers
  9. Keywords:
  10. Quadruped Robot ; Walking ; Two Layer Controller ; Time Invariant ; Learning

 Digital Object List

 Bookmark

No TOC