September Recap @axisrobotics
Binance Wallet campaign
@BinanceWallet × Axis ran for a month (8.26–9.23): 2,521 contributors, 477,000 task completions. Keyless goes directly into Axis—those who participated also got badges.
CEO × Brian Armstrong
@chris_anm01 talked in Singapore with @brian_armstrong
Brian thinks robots will enter the real world within about 7 years; the bottleneck is not hardware but data.
Chris had Brian try the robotic arm tasks in the Axis Hub on-site. No need for physical robots—training data can be generated directly in the browser.
Brian emphasized that more data isn’t necessarily better. Quality and verifiability are key. Every time a robot fails, that failure becomes training material for the next round.
He also did a 22-minute deep podcast with @BasedIndia, covering everything from Chris’s startup journey to the train-to-earn model.
Data Engine
Trajectory data has surpassed 5 million; it’s now over 6 million and still growing, with 200,000+ contributors. The Franka dataset is the most downloaded open-source simulated Franka manipulation dataset on @HuggingFace; KAIST, Tsinghua, and Northwestern universities are using it. π0.5 pretraining improved from 83.9% to 88.8%.
Academia
A paper was accepted by the IROS 2026 Physical World Models workshop and will be presented in Pittsburgh on October 1—top venues in robotics. Embodied RSI success rate increased from 22% to 52%; the model generates its own data to train the next generation.
Open Axis Benchmark
Collaborating with @openroboto, with a 6,000+ task library and 5.5 million+ trajectories
Community
@KaitoAI Phase 2 runs until October 18. The Challenger Program is live—high-difficulty task pools have doubled rewards. Officially confirmed that points will be one of the bases for future airdrops; Season 1 already has 34,686 contributors.
@axisrobotics @KaitoAI
Experience link: http://s.kaito.ai/HdfrZ0t
Binance Wallet campaign
@BinanceWallet × Axis ran for a month (8.26–9.23): 2,521 contributors, 477,000 task completions. Keyless goes directly into Axis—those who participated also got badges.
CEO × Brian Armstrong
@chris_anm01 talked in Singapore with @brian_armstrong
Brian thinks robots will enter the real world within about 7 years; the bottleneck is not hardware but data.
Chris had Brian try the robotic arm tasks in the Axis Hub on-site. No need for physical robots—training data can be generated directly in the browser.
Brian emphasized that more data isn’t necessarily better. Quality and verifiability are key. Every time a robot fails, that failure becomes training material for the next round.
He also did a 22-minute deep podcast with @BasedIndia, covering everything from Chris’s startup journey to the train-to-earn model.
Data Engine
Trajectory data has surpassed 5 million; it’s now over 6 million and still growing, with 200,000+ contributors. The Franka dataset is the most downloaded open-source simulated Franka manipulation dataset on @HuggingFace; KAIST, Tsinghua, and Northwestern universities are using it. π0.5 pretraining improved from 83.9% to 88.8%.
Academia
A paper was accepted by the IROS 2026 Physical World Models workshop and will be presented in Pittsburgh on October 1—top venues in robotics. Embodied RSI success rate increased from 22% to 52%; the model generates its own data to train the next generation.
Open Axis Benchmark
Collaborating with @openroboto, with a 6,000+ task library and 5.5 million+ trajectories
Community
@KaitoAI Phase 2 runs until October 18. The Challenger Program is live—high-difficulty task pools have doubled rewards. Officially confirmed that points will be one of the bases for future airdrops; Season 1 already has 34,686 contributors.
@axisrobotics @KaitoAI
Experience link: http://s.kaito.ai/HdfrZ0t
