Fern launches a simulation platform that needs three robot demos to replace hardware evals
Robert Xu and Amit Yadav's two-person YC startup says three teleop episodes are enough to build a simulation, and one customer has run 900,000 simulations.
By Ryan Merket · Published
Primary source: x.com
Why it matters
Robotics teams burn scarce robot hours on evaluations that simulation could run in parallel. If Fern's learned environments reliably predict real-world outcomes, policy evaluation and RL training could shift from hardware schedules to API calls.
Robert Xu (@robbyuchen), co-founder and CTO of Fern by Ishiki Labs, launched the company's high-fidelity simulation platform for robotics on November 4th in a post on X. Xu said Fern can create a simulation from three teleoperation episodes, then run thousands of rollouts in parallel through an API, with no simulation infrastructure or solvers for customers to manage. He also said one unnamed customer has already run more than 900,000 simulations on the platform.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.