Fern launches scalable robot-policy evaluation platform
Fern launched a managed robot-policy evaluation platform combining learned simulators with access to real bimanual robots.
Evidence notes
- The platform fits action-conditioned world models to customer teleoperation recordings.
- Fern said one model can generate four physically consistent camera views from a robot action stream.
- The launch also offered evaluation on a lab fleet of bimanual 16-degree-of-freedom robots.
Company context
Fern builds learned world-model simulators and reinforcement-learning environments for robot-policy evaluation and improvement. Its platform fits action-conditioned simulations to real robot data and runs policies without physical hardware. In July 2026, Fern reported that a customer achieved 30% more successful completions per robot-hour after deploying its RL model.