Axis Robotics publishes the AXIS community-driven robot-learning data engine
The AXIS paper presents a community-driven robot-learning data engine and benchmark built from 207 tasks and more than 50,000 trajectories.
Why it matters
The AXIS research paper describes a community-driven data engine for robot learning.
Evidence notes
- The released AXIS dataset contains 207 manipulation tasks and more than 50,000 trajectories.
- The system combines browser-based teleoperation with automated task generation, success checking, trajectory cleaning and visual and physics-based augmentation.
- Continual pretraining improved the reported overall success rate of π0.5 by 5.8% and outperformed the RoboCasa365-pretrained comparison by 37.3%.
Company context
Provider of distributed simulation and robot-training data infrastructure.