Browser Platform Crowdsources Robot Training Data from Thousands
Users demonstrate tasks and corrections in browser to generate verified robot training data.
TLDR
Axis Robotics runs a browser platform for robot policy training. Participants complete tasks by controlling virtual arms, after which their actions become trajectories for verification and scoring. The company announced recent gains in policy takeover constraints, cleaner derived tasks, and a second DAgger round alongside domain randomization and LIBERO Pro publication. One early user described failing a rotate-and-place task three times before slowing movements to produce smooth joint paths and accurate contacts. All collection occurs without downloads or external controllers.
Combined views
—
6 Sources, first seen 28d ago
Sources
𝖠𝖥 
𝖠𝖥Δ𝖲𝖸.eth🍌@muhammadafasy20@Yosefphr Exactly. Seeing is data, but understanding is intelligence. The perception → context → action → learning loop is what could make Physical AI truly useful.
CA Catq.eth@catqpx@Shamim_TBs @axisrobotics this approach to data generation might really push the boundaries of how we train AI models, super exciting potential here
FM Farid Mobtadi base.eth BDX@FaridM43370@axisrobotics Great progress this week. The DAgger improvements and sim-to-real setup look especially promising. Excited to see what comes next! 🚀
IM Imran☪️🕊️@ImranTjr@MdRahi444797 @axisrobotics It is much more practical than just relying on perfect demonstrations that fail the moment something changes.
ZI ZIYU@ZIYU_001@axisrobotics 你妈个逼，不会做项目就别做，手机能穿模不说还特么会隐身。