Physically Consistent Online Inertial Adaptation for Humanoid Loco-manipulation

TL;DR

The paper addresses online adaptation of humanoid inertial parameters under significant external loads by formulating an online Kalman-filter-based estimation on a physically consistent parameterization. It introduces a compact decomposition of the inertial regression problem, uses a log-Cholesky mapping to guarantee physical consistency, and integrates the estimator with a whole-body controller for loco-manipulation tasks. Simulation and hardware experiments (e.g., Nadia with a 12 lb load) demonstrate that physically consistent online estimates prevent nonphysical drift and improve task performance, notably under load changes and contact impulses. The approach reduces reliance on fully accurate offline models and enables robust interaction with challenging environments, with innovation gating to mitigate transient estimation spikes.

Abstract

The ability to accomplish manipulation and locomotion tasks in the presence of significant time-varying external loads is a remarkable skill of humans that has yet to be replicated convincingly by humanoid robots. Such an ability will be a key requirement in the environments we envision deploying our robots: dull, dirty, and dangerous. External loads constitute a large model bias, which is typically unaccounted for. In this work, we enable our humanoid robot to engage in loco-manipulation tasks in the presence of significant model bias due to external loads. We propose an online estimation and control framework involving the combination of a physically consistent extended Kalman filter for inertial parameter estimation coupled to a whole-body controller. We showcase our results both in simulation and in hardware, where weights are mounted on Nadia's wrist links as a proxy for engaging in tasks where large external loads are applied to the robot.

Figures (8)

• Figure 1: Our humanoid robot Nadia actively estimating and adapting to 12lb dumbbells attached to the forearms. We use these weights as a proxy for inertially significant manipulation tasks, such as heavy tool use.
• Figure 2: Block diagram of proposed filter and its interface with the controller, as described in Section \ref{['sec:integration']}.
• Figure 3: Parameter estimates for the robot's torso link from the baseline Kalman filter and the Log-Cholesky extended Kalman filter.
• Figure 4: Pelvis height task performance over rough terrain, with and without adaptation of the torso inertial parameters.
• Figure 5: Elbow jointspace task performance with and without adaptation of the forearm link inertial parameters. Around the 8s mark, the mass waveform is changed from a sine wave to the square wave with the same properties.