Learning of Balance Controller Considering Changes in Body State for Musculoskeletal Humanoids

TL;DR

Musculoskeletal humanoids pose balance control challenges due to flexible, redundant bodies and time-varying dynamics. The authors introduce a Deep Predictive Model with Parametric Bias (DPMPB) that learns joint–muscle–$ZMP$ relationships and encodes changing body state into a low-dimensional parametric bias for online adaptation. They collect real-world data from the Musashi humanoid and demonstrate online PB updates enabling adaptive balance control without full-body dynamics modeling, achieving faster convergence and improved stability under disturbances. Across simulation and real-robot experiments, DPMPB enables state-aware balance control robust to changes in posture, calibration, and footwear, highlighting its potential for autonomous control of complex, human-like bodies.

Abstract

The musculoskeletal humanoid is difficult to modelize due to the flexibility and redundancy of its body, whose state can change over time, and so balance control of its legs is challenging. There are some cases where ordinary PID controls may cause instability. In this study, to solve these problems, we propose a method of learning a correlation model among the joint angle, muscle tension, and muscle length of the ankle and the zero moment point to perform balance control. In addition, information on the changing body state is embedded in the model using parametric bias, and the model estimates and adapts to the current body state by learning this information online. This makes it possible to adapt to changes in upper body posture that are not directly taken into account in the model, since it is difficult to learn the complete dynamics of the whole body considering the amount of data and computation. The model can also adapt to changes in body state, such as the change in footwear and change in the joint origin due to recalibration. The effectiveness of this method is verified by a simulation and by using an actual musculoskeletal humanoid, Musashi.

Figures (11)

• Figure 1: The concept of this study.
• Figure 2: The overall system of balance control for musculoskeletal humanoids using a deep predictive model with parametric bias.
• Figure 3: The muscle arrangement of the musculoskeletal humanoid Musashi.
• Figure 4: Simulation experiment: the transition of $\theta^{ref}$ and $z_x$ when conducting Random, Gradual, and Proposed Collections.
• Figure 5: Simulation experiment: the arrangement of parametric bias when training DPMPB using the data collected with Random, Gradual, and Proposed Collection, and the trajectories of parametric bias when running online learning by setting $(\theta_{s-p}, \theta^{offset}_{a-p}) = \{(-5.0, 0.5), (5.0, -0.5)\}$.