Antioch Raises $8.5M Seed for Robot Simulation Tools
Antioch has secured $8.5 million in seed funding at a $60 million valuation to build simulation tools that help robot developers bridge the sim-to-real gap. This gap refers to the difficulty of creating virtual environments realistic enough for robots trained in them to function well in the real world. The round came from lead investors A* and Category Ventures, plus MaC Venture Capital, Abstract, Box Group, and Icehouse Ventures.
Physical AI holds potential for engineers to code physical agents much like digital ones. Robotics faces limits from scarce data in physical environments. Firms often construct mock warehouses for testing or monitor factory lines and gig workers to gather data for deep learning models that control robots.
Company Origins and Team
CEO and cofounder Harry Mellsop launched the New York company in May of last year with four others. Alex Langshur and Michael Calvey joined him after cofounding Transpose, a security and intelligence startup sold to Chainalysis for an undisclosed sum. Collin Schlager came from Google DeepMind, and Colton Swingle from Meta Reality Labs.
Mellsop asked, "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?"
Antioch provides tools that let robot builders create multiple digital copies of their hardware. These connect to simulated sensors producing data identical to real-world inputs. Developers can test rare scenarios, run reinforcement learning, or produce training data.
The firm relies on high-fidelity simulations where physics match reality. Otherwise, real machines controlled by these models could fail. Antioch uses base models from Nvidia, World Labs, and similar providers, then adds specialized libraries for ease of use. Data from various customers helps refine simulations beyond what one physical AI company could achieve alone.
Closing the Sim-to-Real Gap
Mellsop noted, "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."
Executives liken their product to Cursor, the AI coding tool. Better simulations drive efforts at major autonomy firms. Waymo, for instance, employs Google DeepMind's world model to test its self-driving software. This approach cuts data needs for new areas, lowering costs to expand autonomous vehicles.
Stay updated
Get the day's AI and automation news in your inbox. No spam, unsubscribe anytime.
Antioch targets newer companies lacking funds for in-house simulations, physical test sites, or millions of sensor-equipped miles. Building such models demands skills distinct from self-driving tech.
Çağla Kaymaz from Category Ventures said, "What happened with software engineering and LLMs is just starting to happen with physical AI. 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."
Focus Areas and Early Wins
Antioch concentrates on sensor and perception systems key to automated cars, trucks, farm equipment, construction machines, and drones. Broader human-like robots remain distant. Though aimed at startups, early users include large multinationals investing in robotics.
Adrian Macneil, former Cruise executive who built data infrastructure there and founded Foxglove for physical AI data pipelines, invests as an angel. At the Ride.AI conference in San Francisco on Wednesday, he stated, "Simulation is really important when you're trying to build a safety case or dealing with very high-accuracy tasks. It's not possible to drive enough miles in the real world."
Macneil seeks physical AI equivalents to SaaS tools like Github, Stripe, and Twilio. "We need a lot more of the entire toolchain to be available off the shelf," he told reporters.
Mellsop added, "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."
David Mayo from MIT's Computer Science and Artificial Intelligence Laboratory tests LLMs with Antioch's platform. AI models design robots, which get simulated tests, even competing like shoving rivals off platforms. Realistic sandboxes may improve LLM evaluation.
Challenges persist to align digital models with reality. Success would enable data loops like Waymo's, boosting model capabilities monthly. Others must build or acquire such tools to match leaders.