Robust Continuous Motion Strategy Against Muscle Rupture using Online Learning of Redundant Intersensory Networks for Musculoskeletal Humanoids

TL;DR

This work addresses robust motion for musculoskeletal humanoids facing single-muscle rupture by learning a Redundant Musculoskeletal AutoEncoder (RMAE) that captures intersensory relationships among joint angle $\bm{\theta}$, muscle tension $\bm{f}$, and muscle length $\bm{l}$. The approach integrates online learning, anomaly detection, and rupture verification within a unified model to adapt control and state estimation in real time, even as the body changes. Key contributions include modular hardware/software design, rupture-aware online learning with modified loss functions, and extensive simulations and real-robot experiments on Musashi showing improved state estimation and control accuracy under rupture, plus successful continuous motion during tasks. The findings demonstrate a practical pathway toward fault-tolerant biomimetic robotics by leveraging muscle redundancy and online adaptation, enabling sustained operation despite localized hardware failure. $\bm{RMAE}$, its latent space $\bm{z}$, and the online-update mechanisms constitute the core innovations enabling robust, real-time robot performance in uncertain, changeable bodies.

Abstract

Musculoskeletal humanoids have various biomimetic advantages, of which redundant muscle arrangement is one of the most important features. This feature enables variable stiffness control and allows the robot to keep moving its joints even if one of the redundant muscles breaks, but this has been rarely explored. In this study, we construct a neural network that represents the relationship among sensors in the flexible and difficult-to-modelize body of the musculoskeletal humanoid, and by learning this neural network, accurate motions can be achieved. In order to take advantage of the redundancy of muscles, we discuss the use of this network for muscle rupture detection, online update of the intersensory relationship considering the muscle rupture, and body control and state estimation using the muscle rupture information. This study explains a method of constructing a musculoskeletal humanoid that continues to move and perform tasks robustly even when one muscle breaks.

Figures (21)

• Figure 1: The overall concept of this study: anomaly detection for muscle rupture, verification of the muscle rupture state, online learning of intersensory networks considering muscle rupture, and robust continuous motion control using them.
• Figure 2: The basic musculoskeletal structure.
• Figure 3: Overview of software components and flow of this study. The muscle rupture information $\bm{r}$ from Verification (red line) changes each component of online updater, anomaly detector, controller, and state estimator.
• Figure 4: A sensor-driver integrated muscle module asano2015sensordriver and a miniature bone-muscle module kawaharazuka2017forearm for ease of replacement.
• Figure 5: The relationship of sensors, the network structure of RMAE, and the training of RMAE.