Fern
About
About Fern
Fern's simulator advances from a starting image and the same action commands used by a physical robot, generating the observations that a policy consumes on its next step. Its bimanual architecture produces four consistent camera views within one network. This closes the loop between policy decisions and simulated consequences, allowing repeated comparisons from identical starting conditions.
Customers can fit world models to their own robot embodiments, tasks and teleoperation recordings. Fern then uses simulated rollouts to identify failures and train an additional reinforcement-learning model over an existing policy. Its customer work includes a rice-scooping robot, with improvements subsequently tested on physical hardware.
Team
- Amit Yadav - Co-founder
- Robert Xu - Co-founder
Timeline
Events
- 2026-05-11 - Product Launch - Fern launches scalable robot-policy evaluation platform
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.
Source
Relations
Relationships
- Anvil Robotics
Openarm Evaluation Hardware - Fern used Anvil Robotics OpenArm robots to validate its learned simulator and bimanual policy-evaluation workflow.
Context
Comparable profiles
- RoboDK
stack adjacency - Simulation And Test Infrastructure
- Foretellix
stack adjacency - Robot Validation
- MORAI
stack adjacency - Simulation And Test Infrastructure
- dSPACE
stack adjacency - Simulation And Test Infrastructure
- ReSim
stack adjacency - Simulation And Test Infrastructure
- VI-grade
stack adjacency - Simulation And Test Infrastructure
- Fysics
stack adjacency - Simulation And Test Infrastructure
- Simumatik
stack adjacency - Simulation And Test Infrastructure
- Midcentury
stack adjacency - Simulation And Test Infrastructure
- Simantic
stack adjacency - Robot Development Platforms
- Godela
stack adjacency - Robot Development Platforms
- Parallel Domain
stack adjacency - Simulation And Test Infrastructure
Context tags
- Embodied AI world models - Stack Layer - Fern trains action-conditioned world models from real robot data.
- Robotics development platform - Stack Layer - Reviewed July 15, 2026 against the company overview and structured profile evidence: robotics development platform.
- Simulation and Test Infrastructure - Stack Layer - Fern provides learned simulation and managed robot-policy evaluation infrastructure.
- Robot development platforms - Stack Layer - Reviewed July 15, 2026 against the company overview and structured profile evidence: robot development platforms.
- Reinforcement Learning Control - Stack Layer - Fern exposes learned environments and trains RL models on top of customer robot policies.
- Robot Validation - Stack Layer - Reviewed July 15, 2026 against the company overview and structured profile evidence: robot validation.