The vision for physical AI is to empower engineers to program real-world autonomous agents with the same ease and flexibility currently applied to digital systems.
However, this ambition remains largely unfulfilled. Robotics development faces significant hurdles due to a scarcity of data from physical environments. To adequately train their machines, companies often resort to constructing costly mock-up warehouses for testing, while a nascent industry is emerging around monitoring factory lines and gig workers to generate the necessary data for deep learning models that control robots.
An alternative, more scalable approach lies in simulation: highly detailed virtual replicas of real-world settings could furnish roboticists with the extensive data and operational workspaces essential for their advancements.
Antioch, a nascent company specializing in simulation tools for robot developers, aims to bridge what the industry terms the "sim-to-real gap." This critical challenge involves rendering virtual environments sufficiently realistic to ensure that robots trained within them can reliably perform in the physical world.
Antioch CEO and cofounder Harry Mellsop articulated this goal, stating, “How can we do the best possible job reducing that gap, to make simulation feel just like the real world from the perspective of your autonomous system?”
To further this mission, the company announced to TechCrunch today that it has successfully secured an $8.5 million seed funding round, valuing it at $60 million. The round was co-led by venture firms A* and Category Ventures, with additional participation from MaC Venture Capital, Abstract, Box Group, and Icehouse Ventures.
Mellsop established the New York-based firm in May of last year alongside four cofounders. Two of them, Alex Langshur and Michael Calvey, previously collaborated with Mellsop to found Transpose, a security and intelligence startup, which was subsequently sold to Chainalysis for an undisclosed sum. The remaining cofounders, Collin Schlager and Colton Swingle, bring experience from Google DeepMind and Meta Reality Labs, respectively.
The imperative for superior simulation capabilities underpins the strategies of many leading autonomy companies. In the realm of self-driving vehicles, for instance, Waymo leverages Google DeepMind’s advanced world model for testing and evaluating its driving algorithms. Theoretically, this methodology promises to reduce the extensive data collection typically required for deploying Waymo vehicles in new territories, thereby cutting a significant cost in scaling autonomous vehicle technology.
Developing and utilizing such models for robotic testing demands a distinct skill set compared to creating self-driving cars. Antioch endeavors to construct a platform that addresses this need for newer companies lacking the substantial capital to develop these capabilities in-house. These smaller entities often cannot afford to build dedicated physical testing arenas or accumulate millions of miles of sensor-laden vehicle data.
Mellsop observed, “The vast majority of the industry doesn’t use simulation whatsoever, and I think we’re now just really understanding clearly that we need to move faster.”
Antioch executives draw parallels between their product and Cursor, the widely used AI-powered software development tool. Antioch enables robot builders to instantiate multiple digital versions of their hardware, connecting them to simulated sensors that accurately replicate the data a robot's software would receive in a real-world setting. These sophisticated environments empower developers to rigorously test edge cases, conduct reinforcement learning, or generate novel training data.
The efficacy of this approach hinges on the simulation's fidelity. The core challenge lies in ensuring that the simulated physics precisely mirrors reality, preventing malfunctions when the trained model is deployed on a physical machine. Antioch initiates its work using foundational models from providers like Nvidia and World Labs, then develops domain-specific libraries to enhance their usability. Executives emphasize that collaborating with diverse customers provides Antioch with a rich contextual understanding, allowing for a level of simulation refinement that no single physical AI company could achieve independently.
“What happened with software engineering and LLMs is just starting to happen with physical AI,” Çağla Kaymaz, a partner at Category Ventures, shared with TechCrunch. “We do a lot of work on dev tools, and we love that vertical, but the challenges are different. With software, you can have these bad coding tools, and the risk is generally pretty contained to the digital world. In the physical world, the stakes are much higher.”
Antioch's current focus predominantly lies on sensor and perception systems, which constitute the primary demand in autonomous cars and trucks, agricultural and construction machinery, and aerial drones. The broader aspiration for physical AI to power generalized robots capable of replicating complex human tasks remains a more distant goal. While Antioch primarily targets startups, some of its initial collaborations have been with large multinational corporations already making significant investments in robotics.
Adrian Macneil possesses profound expertise in this sector. As an executive at the self-driving startup Cruise, he was instrumental in building the company’s data infrastructure, and in 2021, he founded Foxglove, a company that provides similar data pipelines to physical AI startups. Macneil is now supporting Antioch as an angel investor.
“Simulation is really important when you’re trying to build a safety case or dealing with very high-accuracy tasks,” he remarked at the Ride.AI conference in San Francisco on Wednesday. He added, “It’s not possible to drive enough miles in the real world.”
Macneil envisions the emergence of tools for physical AI akin to those that fueled the SaaS revolution—platforms such as Github, Stripe, and Twilio. He conveyed to TechCrunch, “We need a lot more of the entire toolchain to be available off the shelf.”
Mellsop expressed a strong conviction: “We genuinely all think that anyone building an autonomous system for the real world is going to do so in software primarily in two to three years. It’s the first time you can have autonomous agents iterate on a physical autonomy system, and actually close the feedback loop.”
Experiments in this domain are already underway. David Mayo, a researcher at MIT’s Computer Science and Artificial Intelligence Laboratory, is leveraging Antioch’s platform to evaluate Large Language Models (LLMs). In one particular experiment, Mayo tasks AI models with designing robots, then employs Antioch’s simulator to test their creations. The platform can even facilitate simulated competitions, such as pitting rival bots against each other to push one off a platform. Providing LLMs with such a realistic sandbox could establish a novel paradigm for benchmarking their capabilities.
However, before the widespread adoption of AI engineers in the physical world, substantial work remains to bridge the gap between digital models and real-world performance. If this can be achieved, developers will be able to foster the kind of data flywheel that Macneil identifies as crucial to the success of industry leaders like Waymo, where engineers are increasingly confident that each successive model will surpass the capabilities of its predecessor.
For other companies aspiring to replicate this success, the path forward will involve either developing these essential tools internally or acquiring them from specialized providers.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.