About
Why a robotics network belongs on-chain.
Robotics doesn’t scale like software. A web service ships the moment it compiles; a robot policy has to survive ten-thousand rehearsals in the messy, unforgiving physical world first — and reality has no undo button. The teams building embodied AI told us the same thing over and over: the bottleneck isn’t the robot, it’s the rehearsals, and the GPUs to run them.
The wedge
Closing the sim2real gap is a volume problem — thousands of domain-randomized worlds so a policy generalizes instead of overfitting one hand-tuned scene. That volume is compute, and on a hyperscaler the same H100 costs 3–6× what a decentralized network charges for it. So we built the network: idle GPUs pooled from thousands of operators, scheduled by live price, metered by the GPU-second, and settled on Robinhood Chain — where sub-cent fees make per-second billing and operator rewards actually work.
The loop
Three surfaces, one loop. SimArena is where you describe a scene and drop a robot. Fabric is the metered compute that runs it across the fleet. The open, ROS-compatible runtime takes the exact policy you trained and ships it to real hardware — one binary, sim to silicon. No re-write, no lock-in.
Principles
- Non-custodial. The network reads your balance and meters your usage; it never holds your keys.
- Open runtime. The VLA loop is open and ROS-compatible — bring your own checkpoint.
- Permissionless compute. Anyone with a supported GPU can join the pool and earn.
- Honest metering. Per GPU-second at the live spot price. No seat fee, no minimum.
Xoled is built and run by a small, distributed team of robotics and infrastructure engineers. Want to build, operate a node, or just kick the tires?