Iván Hernández Dalas: Antioch raises funding to bring ‘software speed’ to robot development
Antioch says simulation can help robot developers overcome testing bottlenecks. Source: Antioch
Validating a robot’s behavior in the real world typically means renting physical space, manually staging environments, and resetting hardware between every run, according to Antioch. It’s expensive, slow, and covers only a fraction of the scenarios a system will face in production, said the startup.
Antioch today said it has raised $8.5 million to move development and evaluation of autonomous systems out of the physical world and into cloud-based simulation. The company‘s stated goal is to eliminate the need for hardware and elaborate physical testing.
“Robotics teams are spending weeks staging warehouses and investing millions into test facilities to validate their systems,” said Antioch co-founder Harry Mellsop, who previously worked on Tesla‘s Autopilot team. “Meanwhile, companies like Tesla, Waymo, and Anduril spend hundreds of millions a year on simulation infrastructure to minimize exactly that. We think every autonomy team should have access to that level of tooling.”
Antioch founders draw from security, AI experience
Three of Antioch’s five co-founders, Mellsop, Alex Langshur, and Michael Calvey, previously founded Transpose, a security and intelligence platform acquired by Chainalysis in 2023. Langshur said the national security thread has carried over.
“The only economically viable path to reindustrialization runs through robotics and automation, and scalable testing is the rate-limiting step.” he said.
New York-based Antioch was founded in May 2025. The founding team also included Colton Swingle, previously at Google DeepMind, and Collin Schlager, previously at Meta Reality Labs. The company now has a staff of eight.
“The industries LLMs [large language models] are disrupting — software, professional services, knowledge work — represent maybe $8 trillion of the global economy,” asserted Mellsop. “Manufacturing, logistics, construction, energy, and agriculture represent $50 trillion. AI penetration in those physical industries is basically zero today. The industrial revolution that’s coming in physical AI isn’t going to be a sequel to the LLM revolution; it’s going to dwarf it.”
Co-founders Harry Mellsop, Alex Langshur, Colton Swingle, and Collin Schlager. Source: Antioch
Simulation provides trustworthy testing, says Antioch
Many companies do not want to spend the resources needed to keep up with new tools, models, and providers, even before composing disparate tools into usable workflows, noted Antioch.
The company claimed that the simulation ecosystem has matured to the point where testing and validation are now possible with minimal real-world inputs. Physics and rendering engines such as NVIDIA Isaac Sim and Unreal Engine provide increasingly powerful primitives, and generative techniques like world models from World Labs are making it easier to synthesize realistic environments at scale, said Mellsop.
By sharing physics engines and world models across customers, Antioch ensures that its baseline model is calibrated to the real-world data they collect, he told The Robot Report. This helps all customers have confidence in the models without forcing them to rebuild their infrastructure.
“We’re working with teams within our biggest customer account to model the perception systems that will ultimately go into their robots,” Mellsop said. “The first thing we do with all of our customers is to replicate the existing sets of tests they do in the real world and replicate them 1:1 in Antioch simulation.”
“Now your engineers can run exactly those kinds of physical tests that you were running, but now that it’s on a tablet in the palm of their hands, they can run it as many times as they want, and it’s instantaneous,” he explained. “That slashes costs for these companies and accelerates their engineering and development efforts.”
Editor’s note: At the 2026 Robotics Summit & Expo on May 27 and 28 in Boston, there will be sessions on developing embodied and physical AI. Registration is now open.
AI models can help evaluate edge cases
“Once we’ve done replicated their tests and have built trust in the results of the simulation, we’re then able to think about situations that were previously impossible to actually stage, whether that’s because they were too expensive, too dangerous, or just impractical,” added Mellsop. “How does your robot perform when it’s in the snow or fog? Or maybe the the warehouse is on fire, and you want to test that your safety protocols are still working appropriately.”
“One of the key workflows that I’ve seen some of our customers use is to ask our AI assistant to switch from using one SLAM [simultaneous localization and mapping] algorithm to a more up-to-date version,” he recalled. “In Antioch, your robot is entirely represented as code. Swapping one for the other in simulation is really easy for this agent to do.”
“But then it can go and it can run 1,000 different scenarios and report back what got better, what got worse, what regressions happened,” Mellsop said. “Because you can short-circuit the evaluation loop, it’s ultimately where I think we probably all see robotics going, which is that everything is becoming more composable: the hardware side, the software side.”
Antioch has successfully developed simulation for sensor modalities including lidar, radar, and cameras, said Mellsop. Next, the company plans to address deformables, fluid dynamics, and dexterity modeling, which have been difficult to do in simulation.
It is also looking at upgrading embodied AI perception systems for everything from autonomous vehicles and warehouse planning to smart security and construction site mapping.
Antioch says cloud simulation can handle a much wider range of variance than costly physical testing. Source: Antioch
Investment follows FAANG customers
Antioch said it is already working with Fortune 500 technology and logistics companies, as well as FAANG (Facebook/Meta, Amazon, Apple, Netflix, and Google/Alphabet) engineering teams.
“We spent the first six months heads-down building with a deep conviction about what the product should be,” Mellsop said. “Then in November, we started onboarding these massive customers. We learned a lot from those early engagements, and that was great ratification for us that companies across the board experienced these problems we were working on.”
A* and Category Ventures led Antioch’s latest funding. MaC Venture Capital, Abstract, Box Group, and Icehouse Ventures also participated, as did angel investors including Shyam Sankar, chief technology officer of Palantir, and Adrian Macneil, CEO of Foxglove. The company raised pre-seed funding in December 2025.
“Antioch sits at the intersection of AI and robotics and will unlock the science-fiction future we’ve often dreamed of,” said Bennett Siegel, co-founder and general partner of A*.
“The platform removes the friction from physical-world testing, and enables a new generation of embodied AI startups to scale their inventions globally,” he added. “Companies are currently spending hundreds of millions a year to have a horse in the AI robotics race, but Antioch’s technology reduces the hurdles considerably, giving more businesses a chance to innovate.”
Antioch is looking to scale with its latest funding. “One of the most niche, hardest-to-find, and most important roles for us is what we call a simulation engineer,” noted Mellsop. “These are folks who have real-world experience with the sensors that go onto robots, but they’ve also had some experience of working with simulation engines, whether they’re robot-specific like Isaac or more general like physics and rendering engines. Finding people who’ve got those two skill sets together is really hard.”
Simulation can now help robot developers overcome testing bottlenecks, say Antioch’s founders.
The post Antioch raises funding to bring ‘software speed’ to robot development appeared first on The Robot Report.
View Source
