AXIS: A Growable Community-Driven Data Engine for Scalable Robot Manipulation
AuthorsMengfei Zhao, Dihong Huang, Yikai Tang, Peihao Li, Mingxuan Yan, Ruiqi Zhuang, Yanjia Huang, Jie Wang, Hai Zhai, Tony Zhou, Rui Zhang, Zhexi Luo, Yuchen Huang, Jianfei Yang, Jiachen Li
Resources
AXIS turns community-collected robot demonstrations into a scalable data engine that measurably improves vision-language-action manipulation policies.
Key results
Current manipulation task count in the growable dataset.
Human demonstration trajectories in the current dataset snapshot.
Overall LIBERO-Plus success rate after continual pretraining on AXIS-100%.
Reported overall success-rate improvement from continual pretraining of π0.5 on AXIS.
Reported advantage of AXIS continual pretraining over volume-matched RoboCasa365.
Reduction in mean acceleration after smoothing and resampling.
What the paper found
AXIS, developed by Axis Robotics with contributors from UC Berkeley and Georgia Tech, is a growable robot-manipulation data engine that replaces centralized physical-robot collection with browser-based MuJoCo-WASM teleoperation. Its pipeline uses language-guided task generation, normalized 3D assets, automatic success checkers, trajectory validation, static-segment removal, Savitzky-Golay smoothing, fixed-rate resampling, and IsaacSim-based visual and physics randomization. The current Franka Research 3 dataset contains 207 tasks, 50,129 trajectories, and more than 60K scene variants, organized into reproducible AXIS-25%, AXIS-50%, and AXIS-100% snapshots. In evaluation on LIBERO-Plus, continual pretraining of Physical Intelligence’s π0.5 reaches 88.8% overall success with AXIS-100%, compared with 84.7% and 85.7% for AXIS-25% and AXIS-50%; the paper reports a 5.8% overall improvement and a 37.3% advantage over a volume-matched RoboCasa365 baseline. Refinement reduces mean acceleration by 63.9% and mean jerk by 80.8% relative to raw teleoperation. The strongest robustness gains appear under camera and sensor-noise perturbations, supporting AXIS’s central claim that community collection plus automated curation and simulation augmentation can produce continuously expanding training data, although the current study remains primarily simulation-based with only qualitative real-world Franka rollouts.
Original abstract
Learning effective robot manipulation policies requires diverse, high-quality demonstrations, yet existing data pipelines are often difficult to scale because they rely on specialized hardware, centralized operators, or fixed task suites. We present AXIS, a growable community-driven data engine and benchmark for scalable robot learning, which enables browser-based teleoperation for large-scale demonstration collection, automatically generates and validates new manipulation tasks, and transforms community-collected demonstrations into training-ready data through automated success checking, quality filtering, trajectory smoothing, and visual and physics-based augmentation. The AXIS dataset currently contains 207 diverse tasks and 50K+ trajectories. Meanwhile, AXIS organizes data into task snapshots and evaluates policies with a systematic held-out protocol. We compare vision-language-action (VLA) policies under a unified AXIS evaluation suite and analyze scaling behavior across different data volumes. Continual pretraining on AXIS substantially improves the overall success rate of $π_{0.5}$ by 5.8%, outperforms the model pretrained on RoboCasa365 by 37.3%, and exhibits consistent scaling with increasing data volume, with the largest gains observed under layout, sensor-noise, and camera perturbations.
Read the original paperMore in Robotics
Browse all 50 papers →JAMB: Joint Action-Motion Diffusion for Bimanual Manipulation
Chuyang Xiao, Peilin Meng, David Held
JAMB helps two robot arms coordinate by jointly imagining their future movements and the changing 3D scene before acting.
Rolling-WAM: World Action Models with Rolling Imagination
Yinghua Zhou, Junjie Ye, Yiqi Zhao, Hao Dong, Celina Shiyu Wang, Ruohai Ge, Tingyi Yang, Basile Van Hoorick, Gaurav Sukhatme, Vitor Guizilini, Yue Wang
Rolling-WAM keeps future robot actions partially imagined and refined over time, making world-model-based manipulation replan 4.5 times faster.
Training-free Behavior Cloning
Maximilian Adang, Timothy Chen, Lars Osterberg, Aiden Swann, Mac Schwager
A fast, training-free robot controller reuses and corrects demonstration trajectories to deliver traceable behavior at real-time speeds.