Most Robotic

A robotics research lab pushing what's possible with physical AI

San Francisco startup Instance, a Y Combinator Summer 2026 company, builds an evaluation system that grades robot task attempts from rollout videos. Its first product reads a task description and video, returns a completion verdict, and identifies where a run succeeds or fails; the company says it beats frontier vision-language models in accuracy and latency across more than 10,000 human-labeled episodes, seven robot platforms, and eight benchmarks. MIT classmates Claire Mao and Lucy Cai founded the company.

Jul 2026 · YesPressNews article

Instance, now trading as Most Robotic, is building an autonomous evaluation rig that judges robot-policy rollouts and resets scenes without human intervention. Its success verifier is benchmarked on more than 10,000 human-labeled episodes across seven robot platforms. The public benchmark reports a 0.76 macro success-class F1 across eight held-out test sets, compared with 0.53 for a zero-shot Claude Opus 4.8 baseline. The company is built by MIT computer scientists Claire Mao and Lucy Cai.

Jul 2026 · Y Combinator Launch YCNews article

Instance offers a success detector for robot policies that judges task completion from a task description and an uploaded or recorded rollout video. The product stems from cofounder Lucy Cai’s year of manually labeling whether a humanoid robot completes tasks during robotics research. Instance reports its verifier beats a frontier vision-language model on accuracy at a fraction of the latency, across more than 10,000 held-out human-labeled episodes from seven robot platforms.

Jul 2026 · LinkedIn PulseNews article