Problem and Solution Report: One Robot

The Problem: The Robotics Data Bottleneck

In the field of robotics, training effective action policies (Vision-Language-Action models or VLAs) for complex manipulation tasks is notoriously slow and expensive. Traditional methods rely heavily on manual real-world data collection, where human operators must physically demonstrate tasks or supervise robots for thousands of hours. This "robot time" is a significant bottleneck; it is difficult to scale, prone to hardware wear and tear, and makes it nearly impossible to test edge cases or failure modes safely. Tasks involving soft goods (textiles) or complex geometry (box folding) are particularly difficult to simulate accurately using standard physics engines, forcing teams back into slow real-world testing.

The Solution: World Model-Based Simulation

One Robot addresses this bottleneck by building task-specific simulation environments based on "world models." Unlike traditional simulators that may lack visual or physical realism, One Robot's environments are designed to be both visually and interactionally realistic. By using task-specific data to learn contact dynamics and visual appearance, the platform can generate photo-realistic and physics-accurate rollouts. This allows robotics teams to train and evaluate their policies in a virtual environment that closely mimics the real world, effectively replacing thousands of hours of manual data collection with synthetic trajectories.

Value Proposition

The core value of One Robot's solution lies in its ability to find where action policies fail and generate targeted corrective data. By providing a high-fidelity simulation for "hard manipulation" tasks like textiles and box folding, One Robot enables robotics companies to iterate faster and deploy more reliable systems. The solution moves the training process from the physical world into a scalable digital environment, significantly reducing the cost and time required to bring advanced robotic capabilities to market.

Market and Competitors Report: One Robot

Market Landscape and Trends

One Robot operates within the rapidly evolving warehouse automation and robotics market. This sector is currently characterized by a shift from simple automated guided vehicles (AGVs) to more sophisticated autonomous mobile robots (AMRs) and AI-driven manipulation systems. As labor shortages persist and global supply chains demand higher efficiency, there is a growing trend toward "software-defined robotics," where the intelligence of the robot (its policy) becomes the primary differentiator rather than the hardware itself. This has created a massive demand for high-fidelity simulation tools that can bridge the "sim-to-real" gap.

Target Audience

The company's primary target audience includes robotics startups and enterprise robotics teams focused on complex manipulation. Specifically, they target sectors where traditional automation has struggled, such as textiles, apparel, and intricate packaging (e.g., box folding). These industries require robots to handle non-rigid or variable objects, a task that necessitates the advanced training environments One Robot provides.

Competitive Landscape

The competitive landscape for One Robot can be divided into several categories:

• General Simulation Platforms: Large-scale physics simulators like NVIDIA Isaac Sim or PyBullet. While powerful, these often require significant customization to achieve the level of "interaction realism" One Robot claims for specific manipulation tasks.
• Warehouse Robotics Incumbents: Companies like Locus Robotics, Fetch Robotics (Zebra), and GreyOrange. These firms primarily focus on the hardware and fleet management of AMRs rather than the underlying simulation models for manipulation.
• AI Manipulation Specialists: Newer entrants and research-heavy firms focusing on foundation models for robotics. One Robot differentiates itself by focusing specifically on the simulation environment and world models as a service to other robotics teams, rather than just building their own end-to-end robot.

Competitive Advantages

One Robot's primary advantage is its focus on "interaction realism" and task-specific world models. By specializing in hard manipulation tasks that are traditionally difficult to simulate, they offer a niche value proposition that general-purpose simulators may lack. Furthermore, the founders' experience at Tesla and Google provides them with unique insights into the infrastructure required to train large-scale AI models for physical tasks, giving them a significant technical edge in the market.

Founders and Background Report: One Robot

One Robot was founded in 2025 by Hemanth Sarabu and Elton Shon, two engineers with extensive backgrounds in robotics, machine learning, and large-scale AI infrastructure. The company is part of the Y Combinator Winter 2026 batch and is headquartered in San Francisco. The founding team brings together expertise from some of the most prominent organizations in the robotics and automation sectors, including Tesla, Google, and NASA.

Hemanth Sarabu

Hemanth Sarabu has a deep technical background in building machine learning systems for robots. Prior to founding One Robot, he worked on robot learning and perception at Industrial Next (YC W22). His career includes significant roles at Google, NASA Jet Propulsion Laboratory (JPL), and Symbio Robotics. Additionally, Sarabu has experience in the entrepreneurial space, having bootstrapped Crescer AI, a geospatial AI company, to profitability. His early career also involved building vehicle simulations at McLaren, providing him with a strong foundation in the simulation technologies that One Robot now leverages.

Elton Shon

Elton Shon complements the team with a strong background in robotics and AI infrastructure. He spent five years at Tesla, where he was instrumental in building robotics and large-scale AI infrastructure for factory automation. His tenure at Tesla spanned the critical Model 3 production ramp and work on the Dojo supercomputer team. Following his time at Tesla, Shon joined Industrial Next (YC W22) as the Head of Software, where he focused on developing robot learning and control systems. This experience in high-stakes, large-scale automation environments is central to One Robot's mission of accelerating robot training through advanced simulation.