Learning Whole-Body Quadrupedal Pushing Across Geometry and Physics Variation

Ebasa Temesgen, Dhyan Thakkar, Sarah Boelter, Greta Brown, Maria Gini

Keywords: Mobile Manipulation, Reinforcement Learning, Contact-rich control, Whole-body Loco-manipulation

TL;DR: A hierarchical reinforcement learning pipeline enables robust whole-body quadrupedal pushing across diverse object geometries and physics conditions.

Abstract: Whole-body non-prehensile manipulation allows quadrupedal robots to reposition large objects through contact, but reliable performance remains difficult because pushing depends on object geometry, friction, mass, and intermittent contact dynamics. This paper presents a hierarchical reinforcement learning pipeline in IsaacLab for goal-conditioned whole-body quadrupedal pushing diverse cuboid and cylindrical objects. A high-level policy outputs planar velocity commands, while a frozen low-level locomotion controller provides stable execution. During training, the policy is conditioned on privileged object-context features available in simulation, allowing us to study what a strong simulation pipeline can achieve under substantial geometry and physics variation. We evaluate the learned policy under in-distribution conditions, out-of-distribution geometry, out-of-distribution physics, and on geometries not encountered during training. The results show that the privileged pipeline achieves robust whole-body pushing across objects.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 23