Homepage
Enabling physical AI at scale
fern.botCrawled Sep 27, 2026
Enabling physical AI at scale
Fern shifts from a realtime meeting assistant for sales to creating custom world models for robots. StartupHub.ai describes its learned-simulation approach, which lets teams evaluate and train robot policies without physical hardware.
Axomap surveys 22 startups building world models across physical, virtual, biological, and social systems. It places Fern’s learned robot world models in the physical-AI group.
Reports a customer reinforcement-learning result validated in Fern's world-model simulator: 30% more successful completions per robot-hour and a 7% absolute gain in task success measured on real hardware.
CB Insights analyzes Y Combinator’s Winter 2026 batch, identifying it as the accelerator’s most technically complex cohort yet. Its review of all 199 companies centers on physical AI’s shortage of training data and names Fern among startups that let robotics teams run evaluations in high-fidelity simulation environments built from their own training data.
Robert Xu says his team has closed the loop on robot policy evaluation using Fern's learned world model simulator on a food-prep robot. The update describes a milestone in evaluating robot policies with the simulator, but provides no details about the evaluation method or outcomes.
Announces the launch of Fern, a scalable robotics evaluation platform that trains learned world-model simulators from real teleoperation data.
Y Combinator highlights Fern from Ishiki Labs, a realtime AI meeting assistant that can join meetings as a visible participant or silent copilot. The description presents two ways Fern can take part: joining openly as a participant or operating quietly as a copilot.
Fern is a real-time AI that joins meetings as an invisible copilot or active participant to coach users through conversations. It offers both an unobtrusive mode and a mode in which it actively participates.
Robert Xu says he and Amit Yadav are building Ishiki Labs as part of Y Combinator’s W26 batch. The announcement says the pair left Meta to work on the company, but does not describe Ishiki Labs’ product or market.
Amit Yadav leaves Meta to join Y Combinator and build Ishiki Labs, which focuses on human-like multimodal AI assistants.