Reducing Symmetry Mismatch Caused by Freely Placed Cameras in Robotic Learning
Keywords: Equivariance, Robotics
Abstract: Equivariant policy learning has been shown to solve robotic manipulation tasks with minimal training or demonstration data. However, the effectiveness of equivariance depends on whether transformations of the scene align with simple transformations of the input data. This is true when the camera is in a top-down view, but in the common case where a camera views the robot workspace from the side, there is a symmetry mismatch, reducing model performance. We show that equivariant methods perform better when camera images are transformed to appear as top-down images. Our approach is simple to implement, works for RGB and RGBD images, and reliably improves performance across different view angles and learning algorithms.
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12480
Loading