TI-0.1 Technical Report · Tutor Intelligence

Tutor Intelligence has introduced TRACE, a real-to-sim-to-real agentic harness that uses frontier large language models to reconstruct manipulation tasks in simulation from a few demonstrations. The framework fits appearance and physical parameters using synchronized real-world telemetry, trains privileged-state reinforcement learning policies, and distills them into visuomotor action-chunking policies running on robotic hardware. The company reports high simulation success rates and frames the approach as turning robot learning into a deployment-time compute scaling problem. Open article →