01What is Axis Robotics?
Axis Robotics is building a distributed training-data factory for robots. Contributors teleoperate simulated robots through everyday tasks in a browser, the platform scores each attempt, and the trajectories that pass go into the dataset. When a trained policy fails on a task, the platform opens more collection for that task, so model weaknesses become the next data request.
The platform launched on Base in March 2026 and has grown to over 220,000 contributors and over 7 million trajectories. It spans 13 robot embodiments, from single-arm to humanoid, across kitchen, home, office and workshop scenes.
02How does Axis Robotics collect robot training data?
Through four products that form one pipeline:
- Task generation engine: turns one written instruction into many scene variations; 40 layouts with 4 asset variants give 160 tasks.
- Simulation collection platform: browser-based teleoperation with no robot, GPU or install. MuJoCo runs in the browser; every trajectory is replayed, scored and signed onchain.
- Data-processing pipeline: replay verification, cleaning, domain randomization and photorealistic augmentation.
- Egocentric capture: more than 1,000 trained collectors filming in real homes and businesses; Axis says it holds 200,000+ hours and adds about 5,000 a day.
03Does Axis Robotics data improve robot models?
In the AXIS-V1 paper, accepted at the IROS 2026 Workshop on Physical World Models and co-authored across nine institutions, continual pretraining on Axis data lifted π0.5 from 83.9% to 88.8% on LIBERO-Plus. At equal data volume, Axis data beat RoboCasa365, an established academic simulation dataset, by 37.3 points relative to the baseline (88.8% against 57.5%). Gains were largest under shifted camera angles and sensor noise, and performance scaled with data volume.
These results come from the paper's authors, Axis among them, and cover one robot arm (Franka FR3) in simulation. No independent replication has been published yet.
Axis data on LIBERO-Plus
How often a robot model completed tasks under unseen conditions, by the data it was pretrained on.
π0.5 baselineno extra pretraining
+ Axis dataAXIS-V1, same amount of data
37.3%Axis over RoboCasa365: 31.3 points (88.8% vs 57.5%), measured against the 83.9% baseline.
Success rate on LIBERO-Plus tasks run under conditions the model never saw in training. Same model, same fine-tuning, same evaluation.
04Who are Axis Robotics' customers and partners?
Its first commercial partner is Booster Robotics. Axis rebuilt Booster's workspace as a digital twin, contributors collected 42,000+ simulation episodes, and the resulting model for Booster's T1 humanoid succeeded 87.5% of the time in simulation against 37.5% for an out-of-the-box π0.5. On the physical robot, adding Axis data raised lift success by 5 to 10 percentage points. Booster resells the result to its customers as co-branded task sets.
Axis has also announced partnerships with Feagine Robotics and Dexmal, supplied simulation assets to Manycore Tech, and is working with Lotus and Geely on production-line component sorting covering 500 to 800 SKUs.
05How does Axis Robotics make money?
Axis sells through three channels: embodiment integration for robot hardware companies, robotics data for model teams, and industrial sim-to-real for enterprise deployment. Its whitepaper targets $1M+ in annual recurring revenue by the end of 2026. Axis has told us the framework agreements it is fulfilling exceed $1M in total; that figure has not been independently verified.
06Does Axis Robotics have a token?
Not yet. Axis has published tokenomics for $AXIS, planned on Base, to reward contributors for accepted data and fund partnerships and research. Staking for access to higher-value tasks is proposed but not defined, and nothing yet ties platform revenue to the token. Axis also runs SN/04, a subnet on the BitRobot Network, where contributors earn rewards from both networks.
07Who founded Axis Robotics and who backs it?
Chris Feng (CEO) previously ran data infrastructure as COO of Chainbase. Christine Cao (CMO) held operations roles at Uber, Walmart eCommerce and Nestlé. Professor Jiachen Li of Georgia Tech advises the research. The team is about 30 people, and the $12M seed in July 2026 was led by Hack VC with Nomad Capital, Pi Network Ventures, 10K Ventures and angels.
08What are the main risks?
- Sim-to-real evidence is self-reported: the benchmark results are measured in simulation, and no independent group has yet shown Axis data improving a real robot.
- Single-embodiment evidence: the paper's results all cover the Franka FR3 arm.
- Value accrual undefined: contributors are rewarded in $AXIS, so the incentive depends on the token's price rather than on what buyers pay for data.
09What should we watch next?
- Independent benchmark: a third party confirming or refuting the 37.3-point gap against RoboCasa365.
- Sim Dataset V2 paper: 1.2M trajectories across 1,200 tasks, under review at a leading robotics conference.
- First correction dataset: a large-scale human-gated correction dataset planned before the end of 2026.
- Revenue: converting $1M+ in framework agreements into recurring revenue customers can confirm.
- $AXIS launch: first market pricing of the token.
10Growth
7M+ trajectories in seven months
Cumulative trajectories recorded on Axis since its March 2026 launch on Base.
- 0.19MApr 1
- 0.35MMay 1
- 0.74MJun 1
- 1.23MJul 11
- 2.15MAug 1
- 4.69MSep 12
- 6.87MOct 1
- 7M+Oct 5
1 Axis V2 launched July 102 Booster result published September 4
Contributors: 39,944 on April 1, 221,578 on October 1. Trajectories grew 35x over the same six months.