One Robot
World models for VLA evals and training.
Winter 2026ActiveIndustrialsManufacturing and RoboticsHard TechMachine LearningRoboticsAISan Francisco, CA, USA
Company
https://www.onerobot.ioOne 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.
Verdict
High Signal
Market Opportunity
Robotics simulation infrastructure for VLA training targets a rapidly expanding market as humanoid and industrial manipulation companies (Figure, Physical Intelligence, Apptronik, etc.) scale training pipelines. The ICP is clear: robotics teams building action models for manipulators bottlenecked by real-world data collection. Manufacturing automation alone is a multi-billion dollar market and simulation tooling is becoming a critical infrastructure layer.
High Signal
Founder Signal
Hemanth Sarabu: Head of AI at Industrial Next (YC W22) where he led the pivot to end-to-end robot learning, prior roles at Symbio Robotics (deployed at Ford/Nissan/Toyota), Google, and NASA JPL Graduate Fellowship. Also bootstrapped Crescer AI to profitability — a real exit signal. Elton Shon: employee #2 at Industrial Next as Head of Software Engineering, and 4.5 years at Tesla including early work on Dojo and Autopilot. Both founders worked together at Industrial Next and bring deep, directly relevant robotics ML and systems experience.
Medium Signal
Competition
Competitors include Isaac Sim (NVIDIA), MuJoCo/PyBullet for physics simulation, and newer generative sim startups like Archetype AI and potentially Physical Intelligence's internal tools. One Robot's differentiation is task-specific world models trained on customer data rather than generic physics engines, which is meaningful for contact-rich and deformable object tasks. However no competitor data was returned in research, and this space is heating up fast with well-funded incumbents.
Medium Signal
Product
Website shows a launch video and a 60-second autoregressive world model rollout demo, indicating a working prototype. No named customers, no pricing page, no revenue metrics — just a contact email asking robotics teams to reach out. Product exists but is pre-commercial.
OverallA Tier
Two deeply technical co-founders with overlapping pedigree — both were key builders at Industrial Next (YC W22) working on exactly this problem domain, and Hemanth has a prior bootstrapped-to-profitability company as additional signal. The product has a working demo and a clear, urgent pain point: robotics teams need more simulation fidelity for hard manipulation tasks without burning robot time. No revenue or named customers yet is the main gap, and NVIDIA Isaac Sim is a serious incumbent. But the team's direct industry experience building industrial robot learning systems is rare and credible — they've lived this problem firsthand.
Active Founders
