Hybrid Mode Sample-Based Control

Yilang Liu, Haoxiang You, Ian Abraham

Keywords: Whole-Body Control, Locomotion, Manipulation

TL;DR: We introduce a hybrid mode sampling-based control method for whole-body locomotion and in-hand manipulation

Abstract:

In this work, we propose a novel sample-based approach for manipulation and whole body control problem.
Our method reduces the search space by explicitly accounting for control modes—namely when to apply them, how long to apply them, and which mode to use. We demonstrate that our approach can synthesize complex behaviors, ranging from in-hand manipulation to humanoid locomotion, and achieves significantly better performance in terms of total cost compared to existing methods.
Finally, we validate our method on a physical Unitree A1 platform, highlighting its sample efficiency and real-world applicability.

Submission Number: 2