Fix one policy failure. Measure the improvement.
Bring us one task your robot struggles with. SIM XR reproduces the task in simulation, collects targeted human demonstrations exactly where the policy breaks, validates every trajectory, and measures what changed — without consuming additional robot hours on your side.
Bring us one failure caseSuccess rate on a structurally new manipulation task — before and after fine-tuning on our targeted demonstrations, same evaluation protocol and initial states.
Recorded by a single operator. Baseline: GR00T N1.7, fine-tuned by NVIDIA for its own G1 benchmark task, scored 0% on this structurally new task.
From task definition to a measured policy — collection, fine-tune and evaluation included.
How a sprint runs.
The failing task, success criteria, action/observation schema and evaluation protocol — agreed up front.
The task and its failure distribution in simulation — including your environment, reconstructed from a scan when the scene matters.
Targeted VR demonstrations aimed at the measured failure modes — not blanket hours.
Accept/reject gates and replay validation on every trajectory. Rejects are our cost, not yours.
Before/after evaluation on the same protocol — in your stack where integration permits, or in our evaluation harness.
One task or a bounded failure distribution, at a fixed project price. Success protocol agreed before the first session; demonstrations, validation and the evaluation report always included.
What you receive — and why it works.
- Targeted demonstrations in your schema — LeRobot-ready, replayable, with per-episode logs
- Validation results — accept/reject outcomes and replay checks for every trajectory
- Evaluation report — baseline vs. post-training on identical initial states, plus coverage of the failure distribution
- Optional fine-tuned checkpoint and recommendations for the next iteration
- Demonstrations in the robot’s action space — recorded via teleop, not retargeted from video
- Our own operator pool and infrastructure: consumer headsets, GPU servers in the EU and US, instant scene resets
- No client robot time required for data collection — sessions run off-robot, in simulation
- Compatible with NVIDIA Isaac and GR00T workflows, with LeRobot-format export
Show us where your policy breaks.
Describe the task and where the current policy fails. We’ll come back with a sprint scope: evaluation protocol, collection plan, and a fixed price.
One reply from the team. No mailing list.
