Online Learning of Joint-Muscle Mapping Using Vision in Tendon-driven Musculoskeletal Humanoids

TL;DR

This work addresses the mismatch between geometric models and actual tendon-driven musculoskeletal humanoids by introducing an online neural-network-based joint-muscle mapping (JMM). It derives a smooth muscle Jacobian from the NN, enables online refinement from both estimated states and vision, and maintains stability with a two-updater system (Antagonism Updater and Vision Updater). The approach achieves about a 40% reduction in joint-angle error within 5 minutes and is demonstrated through a can-grasping task, showing practical viability for adapting to state changes and growth. The method advances tendon-driven humanoid control by reducing dependence on perfect geometry and enabling online adaptation using real-world data.

Abstract

The body structures of tendon-driven musculoskeletal humanoids are complex, and accurate modeling is difficult, because they are made by imitating the body structures of human beings. For this reason, we have not been able to move them accurately like ordinary humanoids driven by actuators in each axis, and large internal muscle tension and slack of tendon wires have emerged by the model error between its geometric model and the actual robot. Therefore, we construct a joint-muscle mapping (JMM) using a neural network (NN), which expresses a nonlinear relationship between joint angles and muscle lengths, and aim to move tendon-driven musculoskeletal humanoids accurately by updating the JMM online from data of the actual robot. In this study, the JMM is updated online by using the vision of the robot so that it moves to the correct position (Vision Updater). Also, we execute another update to modify muscle antagonisms correctly (Antagonism Updater). By using these two updaters, the error between the target and actual joint angles decrease to about 40% in 5 minutes, and we show through a manipulation experiment that the tendon-driven musculoskeletal humanoid Kengoro becomes able to move as intended. This novel system can adapt to the state change and growth of robots, because it updates the JMM online successively.

Figures (14)

• Figure 1: Difficulty of movement to the intended position. This figure shows the difference between the movement of a geometric model and the movement of the actual robot when grasping object.
• Figure 2: Overview of the expression of joint-muscle mapping using neural network.
• Figure 3: System overview.
• Figure 4: Joint structure of Kengoro upper limb.
• Figure 5: Muscle arrangement of Kengoro upper limb.