A new startup called Physicl has raised $10 million in a seed round led by Inovia Capital to tackle a longstanding hurdle in robotics: making simulated environments realistic enough to train robots effectively. The goal is to bridge the gap between digital worlds and physical reality so that robots can learn reliably before they’re deployed into the real world.
From 3D shopping visuals to robotic perception
Founder and CEO Alex de Vigan previously built a company focused on helping brands generate 3D models and immersive product renderings. As generative AI advanced, that business shifted toward image synthesis. Now, de Vigan is repurposing his team’s experience and tech pipeline to achieve a different aim: rendering detailed digital environments with accurate geometry, materials, and physics for robotics training.
Robotics researchers have long relied on real-world video and remote operations to train models, but these approaches are costly, slow, and limited—especially when dealing with rare or complex edge cases. Simulations offer a promising alternative, but they often fall short in depicting physical properties and object interaction realistically.
Studio: synthetic data on demand
Physicl’s product, Studio, delivers a library of synthetic training resources built for real-world robotics problems. It includes 10,000 3D environments and 100,000 assets, allowing developers to generate richly varied scenes—for example, cluttered counters or objects partially hidden in shadow—for training robots to perceive and navigate challenging scenarios.
One early user is SpAItial, a world model toolmaker whose offering, Echo-3, converts 2D imagery into 3D simulations. The company has used Physicl’s data to improve realism in its models. SpAItial’s founder says what’s exciting is the potential for this synthetic pipeline to adapt dynamically—detecting where more data is needed and generating it on the fly during model training.
Looking ahead, Physicl envisions customers uploading enormous amounts of data—from images to LIDAR scans—and transforming that raw input into high-fidelity 3D representations. Those representations could then seed billions of synthetic variations that help robots better handle diverse, difficult tasks.
Physicl operates in a crowded and fast-moving space, with many startups trying to supply high-quality data, labeling tools, or simulation platforms. What Physicl is betting on is not just being a data provider but offering the tooling and workflow for ongoing, automated synthetic data generation tailored to specific robotic training needs.
Analytical angle: This raise reflects a broader shift in robotics and AI toward ‘physical AI’—systems that can perceive, reason about, and interact with the real world, not just interpret images or text. For robotics to scale—from warehouses to homes—the fidelity of simulation will matter heavily. What to watch: whether Physicl can keep up with the diversity and complexity of real-world data and whether automatic on-the-fly synthetic data generation becomes a standard part of AI model training pipelines. If Physicl succeeds, it could tilt the balance toward simulation-heavy development, lowering costs and accelerating robotic deployment in the field.
