Data Collection Pilot for Contact-Rich Assembly
How a manipulation team used RCSV data collection to improve task success from 41% to 78% in six weeks.
A startup team used RCSV teleoperation collection, failure replay, and retraining loops to raise success and reduce iteration time.
Challenge
Low consistency in contact tasks
The team saw unstable insertion behavior and poor recovery after contact failures, with no standardized replay flow.
RCSV solution
- Teleop dataset sprint72 hours of multimodal demonstrations with force and pose streams.
- Failure packet pipelineStructured failure events fed directly to retraining queues.
- Weekly benchmark cadenceRegression checks on fixed contact-rich scenarios.
Results in 6 weeks
- Task success: 41% -> 78%
- Policy iteration cycle: 9 days -> 3 days
- Unrecoverable failure rate: down 52%
Plan your own pilot
Share your robot setup and target task, and we will scope collection and evaluation in one plan.







