Online Self-body Image Acquisition Considering Changes in Muscle Routes Caused by Softness of Body Tissue for Tendon-driven Musculoskeletal Humanoids

TL;DR

This work tackles controllability challenges in tendon-driven musculoskeletal humanoids arising from large geometric-model mismatch and tissue-induced muscle-route changes. It defines self-body image as a joint-tension-length mapping and implements two neural-network components, IJMM and MRCM, initialized from a man-made geometry and refined online via Antagonism Updater and Vision Updater. The dual-model framework enables accurate realization of intended joint angles and resilience to external loads, demonstrated on Kengoro through online learning of joint-angle estimation and a robust heavy-object grasping task, with equations such as $l_{target} = f_{ideal}(\boldsymbol{\theta}) + g(\boldsymbol{\theta}, \boldsymbol{T})$ and ${\Delta}\bm{l}_{comp} = -({\alpha}\boldsymbol{T} + {\beta}\boldsymbol{f}_{geo,abs}(\boldsymbol{\theta})\boldsymbol{T})$ guiding initialization. The results show substantial RMSE reductions in angle estimation (to ~2°) and successful posture restoration under load, highlighting the practical impact for real-world tendon-driven systems. Overall, the approach offers a principled way to fuse geometry-based predictions with tissue-induced route changes for reliable, adaptable humanoid control.

Abstract

Tendon-driven musculoskeletal humanoids have many benefits in terms of the flexible spine, multiple degrees of freedom, and variable stiffness. At the same time, because of its body complexity, there are problems in controllability. First, due to the large difference between the actual robot and its geometric model, it cannot move as intended and large internal muscle tension may emerge. Second, movements which do not appear as changes in muscle lengths may emerge, because of the muscle route changes caused by softness of body tissue. To solve these problems, we construct two models: ideal joint-muscle model and muscle-route change model, using a neural network. We initialize these models by a man-made geometric model and update them online using the sensor information of the actual robot. We validate that the tendon-driven musculoskeletal humanoid Kengoro is able to obtain a correct self-body image through several experiments.

Figures (12)

• Figure 1: Problems of musculoskeletal humanoids. Upper figure shows that the actual robot cannot move as intended in a simulation environment. Lower figure shows that there are movements which do not appear as changes in muscle lengths.
• Figure 2: Overview of this system: initial training and online learning.
• Figure 3: Overview of initial training.
• Figure 4: Causes of muscle-route change: muscle wire elongation, structure deformation, foam cover deformation, and interference of muscles.
• Figure 5: Overview of online learning: Antagonism Updater and Vision Updater.