Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Task-specific Self-body Controller Acquisition by Musculoskeletal Humanoids: Application to Pedal Control in Autonomous Driving

Kento Kawaharazuka, Kei Tsuzuki, Shogo Makino, Moritaka Onitsuka, Koki Shinjo, Yuki Asano, Kei Okada, Koji Kawasaki, Masayuki Inaba

TL;DR

The paper addresses the challenge of controlling musculoskeletal humanoids whose self-body image cannot be perfectly aligned with their real dynamics. It introduces DDC-Net, a time-series network that maps sequences of control inputs to predicted time-series task states, enabling real-time task execution by backpropagating loss to refine the initial control sequence. Key contributions include the formulation of a task-specific self-body controller, the detailed network architecture and training/control procedures, and empirical validation on accelerator-pedal control for autonomous driving with the Musashi humanoid, demonstrating faster and more stable convergence than conventional PID methods. This approach has significant implications for enabling autonomous tasks in musculoskeletal platforms without extensive manual controller tuning and suggests avenues for extension to additional actuators and longer-horizon tasks.

Abstract

The musculoskeletal humanoid has many benefits that human beings have, but the modeling of its complex flexible body is difficult. Although we have developed an online acquisition method of the nonlinear relationship between joints and muscles, we could not completely match the actual robot and its self-body image. When realizing a certain task, the direct relationship between the control input and task state needs to be learned. So, we construct a neural network representing the time-series relationship between the control input and task state, and realize the intended task state by applying the network to a real-time control. In this research, we conduct accelerator pedal control experiments as one application, and verify the effectiveness of this study.

Task-specific Self-body Controller Acquisition by Musculoskeletal Humanoids: Application to Pedal Control in Autonomous Driving

TL;DR

The paper addresses the challenge of controlling musculoskeletal humanoids whose self-body image cannot be perfectly aligned with their real dynamics. It introduces DDC-Net, a time-series network that maps sequences of control inputs to predicted time-series task states, enabling real-time task execution by backpropagating loss to refine the initial control sequence. Key contributions include the formulation of a task-specific self-body controller, the detailed network architecture and training/control procedures, and empirical validation on accelerator-pedal control for autonomous driving with the Musashi humanoid, demonstrating faster and more stable convergence than conventional PID methods. This approach has significant implications for enabling autonomous tasks in musculoskeletal platforms without extensive manual controller tuning and suggests avenues for extension to additional actuators and longer-horizon tasks.

Abstract

The musculoskeletal humanoid has many benefits that human beings have, but the modeling of its complex flexible body is difficult. Although we have developed an online acquisition method of the nonlinear relationship between joints and muscles, we could not completely match the actual robot and its self-body image. When realizing a certain task, the direct relationship between the control input and task state needs to be learned. So, we construct a neural network representing the time-series relationship between the control input and task state, and realize the intended task state by applying the network to a real-time control. In this research, we conduct accelerator pedal control experiments as one application, and verify the effectiveness of this study.

Paper Structure

This paper contains 15 sections, 6 equations, 10 figures.

Table of Contents

  1. INTRODUCTION
  2. Musculoskeletal Humanoids, Self-body Image Acquisition, and Autonomous Driving by Humanoids
  3. Musculoskeletal Humanoids
  4. Self-body Image Acquisition
  5. Autonomous Driving by Musculoskeletal Humanoids
  6. Task-specific Self-body Controller Acquisition
  7. Network Structure
  8. Training Phase
  9. Control Phase
  10. Pedal Control Experiments
  11. Experimental Setup
  12. Basic Control
  13. Task-specific Self-body Controller Acquisition
  14. Discussion
  15. CONCLUSION

Figures (10)

  • Figure 1: Autonomous driving by the musculoskeletal humanoid, Musashi.
  • Figure 2: Details of the musculoskeletal humanoid, Musashi.
  • Figure 3: Comparison of robot controls between musculoskeletal humanoids and ordinary axis-driven humanoids.
  • Figure 4: Comparison of autonomous driving between using musculoskeletal humanoids and ordinary axis-driven humanoids. In DARPA robotics challenge darpa2015drc, jaxon kojima2015jaxon needed a jig to sit on the seat.
  • Figure 5: Overview of the task-specific self-body controller acquisition method.
  • ...and 5 more figures