/

Updates

Inside the Lab: SN/03 - TeleArms: RoboPlayground

Play Games, Train Robots

summary

Part 3 of our series exploring the departments inside the BitRobot Lab. This time: SN/03 RoboPlayground on TeleArms, where manipulation tasks become mini-games — and every mission you play becomes training data for real robot arms.

update begins

The department

Every department in the BitRobot Lab attacks the same bottleneck from a different angle. ET Fugi (SN/01) sends players onto real sidewalks. SeeSaw (SN/02) turns everyday phone captures into millions of tasks. SN/03 opens a new front: teleoperation in simulation.

RoboPlayground puts you in control of virtual robot arms inside a high-fidelity physics simulator. Stack blocks, open cabinets, pour liquids, solve multi-step puzzles — with a Joy-Con, keyboard, or full VR. Complete missions, earn rewards, and your gameplay instantly becomes training data for robotics foundation models.

No hardware. No lab. Just play.

Why simulation

We all know the constraint on developing better robotics foundation models is data, but capturing diverse human demonstrations of how to complete real-world tasks can be hard to scale. Remote teleoperation of real robots gets us the richest version of that data, but it hits physical limits: robots are expensive, each serves one operator at a time, and data grows linearly while model demand compounds.

Teleop-in-sim closes the gap. Thousands of people can play simultaneously, tasks can vary procedurally to expose model weaknesses beyond static benchmarks, and the cost per trajectory collapses. Community members have already completed thousands of missions, generating 3000+ hours of rich behavioral data in weeks.

While there are still some trade-offs with the sim-to-real gap, modern simulators, domain randomization, and large-scale human data are closing it faster than ever.

The flywheel

SN/03 isn't replacing real-world teleoperation, it's accelerating it. Simulation data trains policies that transfer to physical robots faster, and the loop compounds: more players → more data → better policies → more exciting missions → more players.

We've already released nearly 200 missions: stack blocks, rotate valves, lift pots, and more. New missions continue to roll out as researchers from the University of Washington and the Allen Institute for AI contribute tasks that reflect real-world manipulation and dexterity. And this is just the beginning. New environments will continue expanding the Playground over time.

Enter the Playground

Whether you're a gamer, researcher, student, or robotics enthusiast, your gameplay matters and early users will be recognized as contributors to the BitRobot network with rewards coming soon through our new unified rewards portal.

Previously in this series: SN/01 – ET Fugi: Catch Aliens, Map the World's Sidewalks · SN/02 – SeeSaw: 4 Million Tasks