Physical AI software · early

Fern

Corporate marketsimulationdata-engineCurated record

fern.bot

Fern (legal entity Ishiki Labs, Inc.) builds learned, action-conditioned world-model simulators for robots, trained end-to-end from a customer's real teleoperation data. Companies upload as little as ~6 hours of teleop recordings per task and Fern fits a high-fidelity world model to that specific embodiment and environment. The simulator is exposed as Gym-style environments for offline/online RL training and as a managed cloud platform for evaluating robot policy checkpoints without running on physical hardware. Because the simulator is learned from the same real-robot data policies train on, the exact deployment checkpoint is evaluated without sim-specific fine-tuning.