Hierarchical Equivariant Policy via Frame Transfer

Haibo Zhao, Dian Wang, Yizhe Zhu, Xupeng Zhu, Owen Lewis Howell, Linfeng Zhao, Yaoyao Qian, Robin Walters, Robert Platt

Keywords: robot learning, imitation learning, robotic manipulation, equivariance

Abstract: Recent work in hierarchical policy learning shows the benefits of splitting control into high- and low-level agents, but current approaches overlook how these levels should interact and often ignore domain symmetries, leading to poor data efficiency. We introduce Hierarchical Equivariant Policy (HEP), which connects the two levels through a frame transfer interface and embeds symmetries directly into both agents. This design provides a strong inductive bias while preserving flexibility, yielding state-of-the-art results in both simulation and real-world manipulation.

Submission Type: Short Research Paper (< 4 Pages)

Submission Number: 59