ir

Hot Keywords

Top

Host

  • Professor Hao Zhang

    Executive Editor of Intelligence & Robotics
    Department of control science and engineering,Tongji University, Shanghai, China.

Speaker(s)

  • Professor Jinhua She

    Associate Editor of Intelligence &Robotics
    School of Engineering, Tokyo University of Technology, Tokyo, Japan.

    Topic: Building a Bridge Between Control Theory and Assistive Robotics

    Japan entered a super-aged society [percentage of old people (≥ 65 yrs old) ≥ 21%] in 2005. The aging problem is very serious in Japan. Studies related to assistive robots have been carried out as measures for the elderly. In this talk, I first present an overview of the current state of rehabilitation and nursing care Equipment in Japan. Then, I'll show the similarity between pedaling and walking exercises using the muscle synergy analysis based on an experiment with a healthy subject. This provides a rationale for the rehabilitation of walking with pedaling exercise. Thus, we developed an electric cart for the elderly. This kind of electric cart installs a foot pedal unit to maintain and enhance the user's walking muscles. Automatic selection of an optimal load of the foot pedal based on the user's physical condition and robust control for various road conditions are realized by incorporating control theories such as H∞ control and dynamic parallel distribution compensation. Finally, I'll explain a left-right-independent lower-limb rehabilitation machine. This device was designed for people with different left and right lower-limb muscle strengths to perform rehabilitation. Some issues, such as hardware design, the division of pedaling area, and a control system design, are discussed.

  • Professor Lei Lei

    Executive Editor of Intelligence & Robotics
    School of Engineering, University of Guelph, Guelph, ON, Canada.

    Topic: Learning Continuous Control with Better Convergence Stability and Performance

    Deep Deterministic Policy Gradient (DDPG) is a popular Deep Reinforcement Learning (DRL) algorithm that has been successfully applied to learn policies over continuous action space for many simulated physics tasks. However, it was found to be less stable in convergence for some scenarios and the performance of the policy can degrade significantly during training, which raise concerns in its practical applications. To improve the stability and performance of DDPG, we present an integrated DRL and Dynamic Programming (DP) approach, where DDPG with fixed target is embedded in a finite-horizon value iteration framework. The approach has been applied for energy management in smart grid, where better performance and convergence stability over DDPG have been demonstrated.

Programme

      Time (9:00, UTC:+8:00)
      Speakers
      Topics
      9:00-9:10
      Hao Zhang
      Welcome remarks
      9:10-9:40
      Jinhua She
      Building a Bridge Between Control Theory and Assistive Robotics
      9:40-10:10
      Lei Lei
      Learning Continuous Control with Better Convergence Stability and Performance
      10:10-10:30
      All
      Q & A Session

Presentation

  • Introduction of Intelligence & Robotics
  • Professor Jinhua She: Building a Bridge Between Control Theory and Assistive Robotics
  • Professor Lei Lei: Learning Continuous Control with Better Convergence Stability and Performance
© 2016-2023 OAE Publishing Inc., except certain content provided by third parties