Online Learning of Danger Avoidance for Complex Structures of Musculoskeletal Humanoids and Its Applications

TL;DR

This work tackles safety in musculoskeletal humanoids by learning a danger probability $p=h_{dan}(\bm{l}^{ref})$ that maps target muscle lengths to risk, enabling preventive control rather than reactive protection. A Danger Avoidance Network (DAN) with online learning, paired with a safety mechanism that relaxes muscle length, is trained initially from geometry-based data and then updated with real-sensor data, achieving improved danger prediction over time (about $72\%$ accuracy after $3\times10^2$ seconds). The authors demonstrate two practical applications: automatic modification of target muscle lengths to safe values and prioritized inverse kinematics to steer away from dangerous postures. This approach reduces dangerous events in a Musashi humanoid and offers a pathway toward integrating safety-aware control into motion planning for complex musculoskeletal systems.

Abstract

The complex structure of musculoskeletal humanoids makes it difficult to model them, and the inter-body interference and high internal muscle force are unavoidable. Although various safety mechanisms have been developed to solve this problem, it is important not only to deal with the dangers when they occur but also to prevent them from happening. In this study, we propose a method to learn a network outputting danger probability corresponding to the muscle length online so that the robot can gradually prevent dangers from occurring. Applications of this network for control are also described. The method is applied to the musculoskeletal humanoid, Musashi, and its effectiveness is verified.

Figures (11)

• Figure 1: The basic structure of musculoskeletal humanoids.
• Figure 2: Various danger situations of musculoskeletal humanoids.
• Figure 3: The whole system of danger avoidance for musculoskeletal humanoids.
• Figure 4: The transition of $f$ and $\Delta{l}_{t}$ when contracting one muscle by 40 mm over 0.5 seconds with a safety mechanism.
• Figure 5: Muscle arrangement of the musculoskeletal humanoid Musashi kawaharazuka2019musashi used in this study.