About
One Robot builds task-specific world-model simulation environments for robotics teams training and evaluating manipulation policies. Its differentiator is realistic visual and interactive simulation learned from robot data, enabling teams to discover policy failures and generate targeted synthetic trajectories without relying as heavily on scarce physical robot time.
Market
One Robot competes in the embodied-AI and robotics simulation market, specifically the infrastructure layer for training and evaluating robot policies and generating synthetic data. Its differentiation is task-specific world models that learn physics and visual appearance from robot data, aiming to deliver both photorealistic and physics-realistic rollouts for difficult manipulation tasks, whereas established simulators are primarily general-purpose physics or synthetic-data platforms.
One Robot targets robotics teams developing manipulators and action models for real-world tasks, particularly in material handling, manufacturing assembly, and other hard-manipulation applications. The likely buyers are robotics, machine-learning, and R&D leaders who need to train and evaluate policies without relying exclusively on scarce physical-robot time.
At a Glance
Problem
Robotics teams currently depend on scarce physical robot time to improve end-to-end action policies: they must set up a scene, run trials, reset the robot, record failures, collect more data, and repeat. That loop is slow, expensive, and difficult to scale, while models need far more training and evaluation data than teams can realistically collect in the real world. The killer use case is hard manipulation in settings such as material handling and manufacturing assembly, including textiles, box folding, contact-rich tasks, and precise assembly where failures are costly and physical testing is especially constrained.
Product / Service
One Robot provides task-specific world-model simulation environments and an evaluation layer for robot manipulation policies. It trains a world model on a customer’s robot data to learn visual appearance and physical behavior, rolls out existing policies in simulation to expose edge cases and failure modes, and generates targeted synthetic trajectories to address those failures. Unlike generic simulators, its models are designed to produce environments that are both realistic to see and realistic to interact with, enabling teams to run more experiments and evaluations, reduce dependence on robot availability, and move reliable policies toward production faster.
Market
One Robot competes in robotics software infrastructure, particularly simulation, synthetic-data generation, world models, and evaluation tools for physical-AI and manipulation systems. Its alternatives include general-purpose robotics simulators such as NVIDIA Isaac Sim, which supports designing, simulating, testing, and training AI-based robots in physically based virtual environments, and open-source platforms such as Genesis, which combines a multiphysics engine with photorealistic rendering. One Robot’s differentiation is learned, task-specific simulation aimed at closing the gap between visual realism and contact-physics realism for difficult manipulation policies.
The company is very early stage: Y Combinator lists it as an active Winter 2026 company founded in 2025 with a two-person team, and its hiring materials say it is backed by Accel, YC, exited founders, and AI engineering leaders. Its materials also say it works with real customer data, real failures, and real deployment pressure, but they disclose no named customers, revenue, or commercial metrics. The best-supported characterization is therefore an early commercial-stage startup with customer-development activity, not a company with publicly demonstrated revenue or scale.
Founders & Leadership
Funding History
Y Combinator
Recent News
StartupHub.ai profiled One Robot, a YC W2026 startup building task-specific world-model simulations from customers’ robot data. The product is designed to let robotics teams test and train vision-language-action policies, identify failure modes, and generate targeted synthetic training data without relying exclusively on physical hardware.
Active Roles
4Business Model
One Robot appears to use a direct B2B model, working with robotics teams that are building action models for manipulators. Its public site invites prospective customers to contact the founders, but does not disclose pricing, subscription terms, or other specific revenue mechanics.