Browse, replay, and teleoperate SO-100 motion datasets. The exchange where roboticists share the demonstrations that teach machines to move.
LLMs had the internet. Robotics has nothing equivalent. Every lab collects its own teleoperation data, in isolation, using expensive hardware and in-person operators.
The result: the best robot learning algorithms are starved of the data they need. Foundation models like RT-X, GR00T, and pi0 are only as good as their training sets. And right now, those sets are tiny.
Embodex changes the equation. A shared motion exchange where every SO-100 dataset is available to every researcher. Collect once, train everywhere.
Explore motion datasets by task, quality score, and complexity. Replay trajectories in 3D before committing to a download. Know exactly what you're training on.
Control an SO-100 directly from your browser. No leader arm required. Contribute demonstrations from anywhere in the world, lowering the barrier to data collection.
Every trajectory is automatically evaluated for smoothness, task completion, and consistency. Bad data gets filtered. Good data gets surfaced. Your models train on the best.
Share your datasets with the community or keep them private. Trade demonstrations across labs. The more people contribute, the better everyone's models get.
Thousands of researchers build with the SO-100. Embodex is where their data comes together, so every robot arm learns from every demonstration ever recorded.