Online Learning Feedback Control Considering Hysteresis for Musculoskeletal Structures

TL;DR

The paper tackles hysteresis in musculoskeletal humanoid control by introducing Online Learning Feedback Control (OLFC) that online-updates a neural network modeling the relationship between joint-angle error and changes in target muscle lengths. It compares two network formulations (Type A and Type B) and multiple loss definitions, demonstrating that a transition-focused Type B network yields faster, more robust convergence, and that loss choices can trade off accuracy against internal muscle tension. Through extensive experiments on the Musashi robot, including varied postures, loads, and a stamp-pushing task with visual feedback, the approach shows generalization across situations and potential to operate with vision-based joint-angle estimation. The work advances practical, hysteresis-aware control for complex, closed-chain musculoskeletal systems, enabling faster stabilization with fewer feedback trials and safer actuation, while outlining avenues for dynamic extension and sensorless operation.

Abstract

While the musculoskeletal humanoid has various biomimetic benefits, its complex modeling is difficult, and many learning control methods have been developed. However, for the actual robot, the hysteresis of its joint angle tracking is still an obstacle, and realizing target posture quickly and accurately has been difficult. Therefore, we develop a feedback control method considering the hysteresis. To solve the problem in feedback controls caused by the closed-link structure of the musculoskeletal body, we update a neural network representing the relationship between the error of joint angles and the change in target muscle lengths online, and realize target joint angles accurately in a few trials. We compare the performance of several configurations with various network structures and loss definitions, and verify the effectiveness of this study on an actual musculoskeletal humanoid, Musashi.

Figures (12)

• Figure 1: Basic structure of the musculoskeletal humanoid.
• Figure 2: A musculoskeletal humanoid Musashi kawaharazuka2019musashi used in this study and the muscle arrangement of its left arm.
• Figure 3: Characteristics of hysteresis in the musculoskeletal structure.
• Figure 4: Evaluation experiment of the basic feedback control using a network of kawaharazuka2018bodyimage. This graph includes the results before and after the training of the network using the actual robot sensor information.
• Figure 5: Network structures for online learning feedback control.