3D Equivariant Visuomotor Policy Learning via Spherical Projection

Boce Hu (Northeastern University) · Dian Wang (Stanford University) · David Klee (Northeastern University) · Heng Tian (Northeastern University) · Xupeng Zhu (Northeastern University) · Haojie Huang (Northeastern University) · Robert Platt (Northeastern University) · Robin Walters (Northeastern University)

$\mathrm{so}(3)$data efficiency diffusion policyequivariant modelseye-in-hand rgb camerafeature projectionmonocular rgb inputspoint cloud inputspolicy learning real-world experiments robotic manipulation sample efficiency simulation experiments strong baselines symmetries

Equivariant models have recently been shown to improve the data efficiency of diffusion policy by a significant margin. However, prior work that explored this direction focused primarily on point cloud inputs generated by multiple cameras fixed in the workspace. This type of point cloud input is not compatible with the now-common setting where the primary input modality is an eye-in-hand RGB camera like a GoPro. This paper closes this gap by incorporating into the diffusion policy model a process that projects features from the 2D RGB camera image onto a sphere. This enables us to reason about symmetries in $\mathrm{SO}(3)$ without explicitly reconstructing a point cloud. We perform extensive experiments in both simulation and the real world that demonstrate that our method consistently outperforms strong baselines in terms of both performance and sample efficiency. Our work, $\textbf{Image-to-Sphere Policy}$ ($\textbf{ISP}$), is the first $\mathrm{SO}(3)$-equivariant policy learning framework for robotic manipulation that works using only monocular RGB inputs.