Teehoo Embodied · Robot data CRO
Every pilot ships a sim-ready scene, a 10× augmented LeRobot v2 trajectory dataset, and a fine-tuned LoRA adapter — drop-in for OpenVLA, π0, RT-X, Octo, and RDT. Four-to-six week engagement. PIPL-audited end-to-end. Delivered onshore by design.
The problem
AgiBot World, Open X-Embodiment, and LeRobot have made generic robot data almost free. Backbone policies pretrained on them already cover a huge surface. None of them, however, cover your specific task on your specific robot at your specific factory.
1M+ household trajectories on someone else's humanoid. Useful backbone. Does not contain a single demo of your SO-100 pouring your customer's reagent into your customer's vial.
Capture rigs, teleop operators, scene reconstruction, augmentation, cross-gripper retargeting, adapter training, and a compliance audit chain. Eight engineering disciplines, one quarter minimum.
Sending Chinese-collected video to an overseas annotation desk invokes formal cross-border data review under PIPL. Slower, costlier, and the auditable lineage becomes somebody else's responsibility.
What you get
Every pilot produces the same three artifacts. Each is a standalone deliverable; together they let your team simulate, train, and ship a new policy on day one of the bundle landing.
Run simulation, evaluation, and policy testing before touching a real robot. Geometry, mass, and friction tuned to your actual environment so the sim and the real workcell agree.
Formats: URDF + MJCF · drop-in for Isaac Sim, Genesis, SAPIEN, MuJoCo
Real on-site demonstrations grown via jitter, time-stretch, pose-offset, and Blender re-rendering. Cross-gripper retargeted across Franka Panda, SO-100, and Allegro Hand so the dataset isn't locked to one embodiment.
Format: LeRobot v2 + Open X-Embodiment compatible · HuggingFace dataset card included
Load into your robotics stack and start evaluation immediately. Fine-tuned on your choice of backbone — OpenVLA-7B, π0, RT-X, Octo, or RDT — trained onshore and delivered onshore-only.
Format: HuggingFace PEFT · backbone choice at pilot kickoff
How it's made
Every Teehoo pilot runs the same six-stage pipeline. Each stage emits an audit event into the bundle's hash chain. You can verify the chain after delivery with one CLI command.
Onshore capture — video, teleop, hand demos.
Real workcell → URDF + MJCF digital twin.
10× via jitter, time-stretch, pose-offset, Blender.
Remapped across Franka, SO-100, Allegro.
LoRA on OpenVLA / π0 / RT-X / Octo / RDT.
tar.gz · dataset card · SHA-256 audit chain.
Stage outputs and per-stage backend modes are visible inside every delivery's Honesty Matrix tab — REAL, MOCK (deterministic stub), or FIXTURE (bundled sample) per stage, with no hand-waving.
PIPL audit chain · onshore by design
Teehoo runs the data ops inside the PRC and delivers the trained adapter onshore-only. Every data-handling event is SHA-256 hash-chained against the previous one — verifiable from your laptop with the bundled teehoo-pipl verify command.
The audit chain is structural verification, not a third-party PIPL compliance certificate. External attestation is a Stage 5 roadmap item.
Source video, intermediate artifacts, and trained adapter weights never leave PRC borders.
Every event references the SHA-256 of the previous one. Tamper with one, the chain breaks visibly.
Every event references the pilot's lawful basis identifier so your counsel can inspect the full ledger.
See it on a real delivery
A four-episode demonstration of an actual delivery: the three-piece bundle, the HuggingFace dataset card, the PIPL audit chain, and the single-command reproduction recipe. No signup; we'll put you in as the demo customer.
