We're building the trust layer for embodied AI — in the open.
Mission
OpenBot exists to close the gap between lab performance and real-world deployment in embodied AI. We believe the bottleneck isn't better models — it's better evaluation, better data, and better sim-to-real transfer.
We're building four pieces of that layer: Bench for policy evaluation, Data for teleop curation, Synth for data synthesis, and an API that ties them together. The public code is available on GitHub, and the managed workflow is in active development with early teams.
What we're not
We don't build robots. We don't train policies for production deployment. We're not a general computer vision platform — detection, segmentation, and OCR are well-served elsewhere.
Our focus is embodied AI: arms, humanoids, mobile manipulators. The metrics that matter are task success, intervention rate, and sim-to-real gap — not mAP or IoU.
Research
We publish papers, notes, and reports on policy evaluation, teleop data quality, sim-to-real transfer, and cross-embodiment generalization. See the research index for recent work.
Our research is applied: the methods and metrics we write about are the ones we're building into the open core — not a separate track.
Company
OpenBot publishes local evaluation, dataset checks, schemas, adapters, and early SDK work on GitHub. Managed rollouts, artifact storage, API keys, and shared workspaces are being developed with early access teams.
For early access and enterprise inquiries, contact contact@openbot.ai.