Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Long-time Self-body Image Acquisition and its Application to the Control of Musculoskeletal Structures

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

TL;DR

This work tackles the control challenges of a tendon-driven musculoskeletal humanoid by introducing a long-time self-body image that decomposes the joint-muscle mapping into two networks: the Ideal Joint-Muscle Mapping (IJMM) and the Muscle-Route Change Model (MRCM). The approach enables stable online learning, data augmentation, and a safety mechanism to limit muscle tension and temperature, enabling three control modalities—position, torque, and variable stiffness—driven by the learned self-image. Empirical results show gradual improvement in joint-angle tracking over hours, successful dumbbell lifting, and meaningful variable stiffness behavior with respect to environmental interactions and impacts. The work suggests practical pathways for extending musculoskeletal robots’ range of applications, including hands and tensegrity-soft systems, through robust self-body image acquisition and control strategies.

Abstract

The tendon-driven musculoskeletal humanoid has many benefits that human beings have, but the modeling of its complex muscle and bone structures is difficult and conventional model-based controls cannot realize intended movements. Therefore, a learning control mechanism that acquires nonlinear relationships between joint angles, muscle tensions, and muscle lengths from the actual robot is necessary. In this study, we propose a system which runs the learning control mechanism for a long time to keep the self-body image of the musculoskeletal humanoid correct at all times. Also, we show that the musculoskeletal humanoid can conduct position control, torque control, and variable stiffness control using this self-body image. We conduct a long-time self-body image acquisition experiment lasting 3 hours, evaluate variable stiffness control using the self-body image, etc., and discuss the superiority and practicality of the self-body image acquisition of musculoskeletal structures, comprehensively.

Long-time Self-body Image Acquisition and its Application to the Control of Musculoskeletal Structures

TL;DR

This work tackles the control challenges of a tendon-driven musculoskeletal humanoid by introducing a long-time self-body image that decomposes the joint-muscle mapping into two networks: the Ideal Joint-Muscle Mapping (IJMM) and the Muscle-Route Change Model (MRCM). The approach enables stable online learning, data augmentation, and a safety mechanism to limit muscle tension and temperature, enabling three control modalities—position, torque, and variable stiffness—driven by the learned self-image. Empirical results show gradual improvement in joint-angle tracking over hours, successful dumbbell lifting, and meaningful variable stiffness behavior with respect to environmental interactions and impacts. The work suggests practical pathways for extending musculoskeletal robots’ range of applications, including hands and tensegrity-soft systems, through robust self-body image acquisition and control strategies.

Abstract

The tendon-driven musculoskeletal humanoid has many benefits that human beings have, but the modeling of its complex muscle and bone structures is difficult and conventional model-based controls cannot realize intended movements. Therefore, a learning control mechanism that acquires nonlinear relationships between joint angles, muscle tensions, and muscle lengths from the actual robot is necessary. In this study, we propose a system which runs the learning control mechanism for a long time to keep the self-body image of the musculoskeletal humanoid correct at all times. Also, we show that the musculoskeletal humanoid can conduct position control, torque control, and variable stiffness control using this self-body image. We conduct a long-time self-body image acquisition experiment lasting 3 hours, evaluate variable stiffness control using the self-body image, etc., and discuss the superiority and practicality of the self-body image acquisition of musculoskeletal structures, comprehensively.
Paper Structure (23 sections, 6 equations, 13 figures)

This paper contains 23 sections, 6 equations, 13 figures.

Table of Contents

  1. INTRODUCTION
  2. Musculoskeletal Humanoid
  3. Joint Structure
  4. Muscle Configuration
  5. Long-time Self-body Image Acquisition
  6. Overview of Self-body Image Acquisition
  7. Comparison of Self-body Images
  8. Initial Training of Self-body Image Using a Geometric Model
  9. Online Learning of Self-body Image Using the Actual Robot
  10. Safety Mechanism Considering Muscle Tension and Temperature
  11. Position, Torque, and Variable Stiffness Control Using Self-body Image
  12. Position Control
  13. Torque Control
  14. Variable Stiffness Control
  15. Estimation of Self-body Stiffness
  16. ...and 8 more sections

Figures (13)

  • Figure 1: Overview of newly developed MusashiLarm and Musashi used in this study kawaharazuka2018musashilarm-en.
  • Figure 2: Overview of self-body image acquisition.
  • Figure 3: Difference of the network configuration between previous study kawaharazuka2018bodyimage and this study.
  • Figure 4: Data accumulation and augmentation of the actual robot sensor information for generation of minibatch for the stable online learning of self-body image.
  • Figure 5: Estimation of operational hardware stiffness using self-body image.
  • ...and 8 more figures