Prof. Robot: Differentiable Robot Rendering Without Static and Self-Collisions

Quanyuan Ruan, Jiabao Lei, Wenhao Yuan, Yanglin Zhang, Dekun Lu, Guiliang Liu, Kui Jia · 2025

Read Prof. Robot: Differentiable Robot Rendering Without Static and Self-Collisions by Quanyuan Ruan, Jiabao Lei, Wenhao Yuan, Yanglin Zhang, Dekun Lu, Guiliang Liu, Kui Jia (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-11269.txt.

Description

Quanyuan Ruan and six co-authors added collision sense to differentiable robot rendering in twenty twenty-five. Such rendering lets robots learn acts from image feedback. But the optimizer has no sense of the real world. It can plan moves that crash the robot into a wall, or into itself. Their fix trains a neural collision classifier with the renderer.

The classifier flags static hits and self-hits. It steers gradient-based action optimization clear of them. One snag was gradients from the classifier that jumped about. Eikonal regularization fixed that and smoothed the guide. The module plugs into current differentiable rendering frameworks. It does so with no need to rebuild them. Tests in words and in numbers show the method is needed. They show gains over past fixes too. The work points to safer robot learning from images.

Catalog record

CREATORQuanyuan Ruan, Jiabao Lei, Wenhao Yuan, Yanglin Zhang, Dekun Lu, Guiliang Liu, Kui Jia
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-11269

Browse the full library · About the Tower of Babel