Musculoskeletal AutoEncoder: A Unified Online Acquisition Method of Intersensory Networks for State Estimation, Control, and Simulation of Musculoskeletal Humanoids

TL;DR

The paper tackles the modeling complexity of musculoskeletal humanoids by introducing the Musculoskeletal AutoEncoder (MAE), a unified online framework that couples state estimation, control, and simulation through intersensory networks among joint angles $\bm{\theta}$, muscle lengths $\bm{l}$, and muscle tensions $\bm{T}$. MAE is trained initially from a geometric model and then updated online using real sensor data, enabling improved state estimation, lower-tension control via backpropagation in latent space, and a more faithful real-time simulator. Key contributions include a structured MAE architecture with latent space manipulations, an online learning routine with masked inputs, anomaly detection based on reconstruction, and demonstration on the Musashi humanoid showing quantitative gains in estimation accuracy and simulator fidelity. The approach advances practical deployment of musculoskeletal robots by integrating estimation, control, and simulation within a single, adaptable model that can continually align with physical hardware.

Abstract

While the musculoskeletal humanoid has various biomimetic benefits, the modeling of its complex structure is difficult, and many learning-based systems have been developed so far. There are various methods, such as control methods using acquired relationships between joints and muscles represented by a data table or neural network, and state estimation methods using Extended Kalman Filter or table search. In this study, we construct a Musculoskeletal AutoEncoder representing the relationship among joint angles, muscle tensions, and muscle lengths, and propose a unified method of state estimation, control, and simulation of musculoskeletal humanoids using it. By updating the Musculoskeletal AutoEncoder online using the actual robot sensor information, we can continuously conduct more accurate state estimation, control, and simulation than before the online learning. We conducted several experiments using the musculoskeletal humanoid Musashi, and verified the effectiveness of this study.

Figures (13)

• Figure 1: Classification of previous studies and positioning of this study.
• Figure 2: The basic structure of musculoskeletal humanoids covered in this study, and the musculoskeletal humanoid Musashi used in this study.
• Figure 3: The network structure of Musculoskeletal AutoEncoder.
• Figure 4: State estimation, control, and simulation methods using Musculoskeletal AutoEncoder.
• Figure 5: The transition of the error between the current and estimated joint angles while executing online learning of Musculoskeletal AutoEncoder.