Ishiki Labs Analysis
What is Ishiki Labs?
Fern builds learned world models that simulate robot behavior from real teleoperation data. Its system runs closed-loop evaluation and reinforcement-learning environments for robotics teams, with 256 by 256 RGB rollouts and multi-camera consistency. The company also offers custom world models and public benchmarks for robot policy developers and research teams.
Location
Brooklyn, United States
Employees
1-10
Founded
2025
Industry
B2B, Infrastructure
Business Model v4
Hardware
Delivery Model
Developer platform / API / infra
Product Description
Fern specializes in creating learned world models that simulate robot behavior based on real teleoperation data. Their main product offerings include:
-
Learned World Models: These models are designed to accurately replicate the behavior of robots in various environments, allowing for more effective training and evaluation.
-
Closed-Loop Evaluation Environments: Fern provides systems that enable robotics teams to conduct closed-loop evaluations, ensuring that robots can learn and adapt in real-time based on their interactions with the environment.
-
Reinforcement-Learning Environments: The company offers environments specifically tailored for reinforcement learning, which is crucial for developing advanced robotic policies. These environments support 256 by 256 RGB rollouts and maintain multi-camera consistency, enhancing the training process.
-
Custom World Models: Fern also develops custom world models tailored to the specific needs of their clients, allowing for more personalized and effective robotic training solutions.
-
Public Benchmarks: They provide public benchmarks for robot policy developers and research teams, facilitating collaboration and standardization in the field of robotics.
Key Features and Benefits:
- Enhanced Training: The learned world models allow for more realistic simulations, leading to better-trained robots.
- Real-Time Adaptation: Closed-loop evaluations enable robots to adapt their behavior based on immediate feedback, improving performance.
- Scalability: The evaluation environments are scalable, making them suitable for various robotics applications and team sizes.
- Customization: Custom world models ensure that specific requirements of different robotics projects are met.
- Collaboration: Public benchmarks promote collaboration among researchers and developers, advancing the field of robotics.
Overall, Fern's offerings are designed to support robot policy developers and reinforcement-learning researchers, providing them with the tools necessary to advance robotic capabilities.
See something that needs updating? Suggest edits to this profile.