Reproduce “AXIS: A Growable Community-Driven Data Engine for Scalable Robot Manipulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mengfei Zhao, Dihong Huang, Yikai Tang, Peihao Li, Mingxuan Yan, Ruiqi Zhuang, Yanjia Huang, Jie Wang and 7 more
AXIS, a growable community-driven data engine from Axis Robotics and collaborating universities, leverages browser-based teleoperation and automated data augmentation to create a vast repository of robot manipulation demonstrations. This infrastructure generated over 50,000 human trajectories for 207 tasks, enhancing downstream vision-language-action policy robustness by 5.8% on the LIBERO-Plus benchmark and exhibiting significant gains against visual and sensor perturbations.
Key insights
- Democratizing robot data collection through a browser-based teleoperation system significantly lowers barriers to entry, enabling large-scale contributions of diverse human demonstrations.
- A systematic data refinement pipeline, coupled with extensive visual and physics augmentation, effectively transforms raw, noisy teleoperated data into high-quality, diverse training examples for robust policy learning.
- Continual pretraining on a progressively growing and diverse dataset like AXIS measurably improves the generalization and robustness of Vision-Language-Action (VLA) policies against various perturbations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.