platoseedplatoseed
World models for robot evals and training.
One Robot builds simulation environments that are realistic to see and realistic to interact with, so robotics teams can train and evaluate robot policies without being bottlenecked by robot time. Today, improving a VLA often means more real-world hours: setting up the scene, running trials, resetting, and repeating. This loop is slow, expensive, and hard to scale. For example, material handling and manufacturing assembly tasks, models need far more training and evaluation data than teams can collect in the real world. We use task-specific data to build world model-based simulation environments for hard manipulation tasks (for example, textiles and box folding). These environments help teams run more training and evals, find failure modes faster, and accelerate iteration on action policies with less dependence on real-world data collection and robot availability.
One Robot builds simulation environments (world models) that are visually and physically realistic, enabling robotics teams to train policies faster by testing edge cases and generating synthetic training data before physical deployment. The focus is on hard manipulation tasks where physics-accurate simulation is critical.
One Robot provides a three-step workflow: (1) Train a world model on your robot's real-world data to learn physics and visuals; (2) Roll out your policy in simulation to discover edge cases before deployment; (3) Generate targeted synthetic trajectories to fix discovered failures. The world models handle contact-rich manipulation, deformable objects, and precise assembly tasks, rendering camera streams conditioned on controller actions.
Who it’s for: Robotics teams building manipulation-focused systems (pick-and-place, assembly, deformable object handling) that need to reduce iteration cycles and test policies before physical deployment.
Founders have backgrounds at Google, Tesla, Industrial Next (YC W22), NASA JPL, and McLaren; actively seeking robotics teams building manipulators for production partnerships.
Bringing robots to life using world models and machine learning. Previously built robot learning and perception systems at Industrial Next, Symbio Robotics, NASA JPL, and Google. Bootstrapped geospatial AI company, Crescer AI, to profitability.
Robotics, SW/FW. Previously built robot learning and control system at Industrial Next and helped build Dojo at Tesla.
Building world model-based environments for robot training and evaluation
One Robot builds world-model–based simulations that learn contact dynamics and visual appearance from robot data to create photorealistic, physics-faithful environments for training and evaluating robot action models, enabling more iterations with less real-world robot time. Target users are robotics teams training manipulators for tasks like textiles and box folding.
Documenting human dexterity
Multimodal data provider for robotics and world modeling
