There is a moment in every robotics lab where someone spends three months wiring a robot arm to a camera, writing inverse kinematics solvers from scratch, and building a teleoperation system by hand before they can finally start teaching the thing something useful. Mike Xia and Vijay Pradeep have seen that moment too many times. Their company, Anvil Robotics, just raised $5.5 million in seed funding to eliminate it.
Anvil, based in San Francisco, sells what Xia and Pradeep describe as Legos for Physical AI teams: modular devkits that bundle the hardware and software a robotics researcher needs to go from unboxing to first training run in a day. The kit for a Physical Intelligence-style arm starts at $5,300. A full OpenARM setup with Quest 3s teleoperation support runs $9,300. Every component ships pre-integrated, the company says, from the 60-frames-per-second camera module to the 500 to 1,000Hz control bus to the inverse kinematics solvers.
The comparison to Lego is deliberate. Anvil's products are designed to snap together and come apart. The same software stack runs on OpenARM and the newer OpenYAM arm. You can swap grippers, lenses, and base units. Most of the hardware is open source. "You're not limited in any way," Anvil writes on its about page. "We'll help you fork and customize any of our robots."
The $5.5 million seed round was first reported by GNews and confirmed by Crunchbase. The company counts its customers as its reference investors: teams inside Nvidia, Google, Co.bot, Path Robotics, and Ultra Tech are already using Anvil hardware. The company was founded by Xia, who was previously chief product officer at Voltage Park, a deep tech investment firm, and Pradeep, whose background includes robotics middleware research published at the Journal of Software Engineering for Robotics going back to 2011. Haomiao Huang, a founding partner at Matter Venture Partners, sits on the board, according to Crunchbase.
Anvil's pitch is partly a cost story and partly a time story. The company estimates that the average robotics team spends months rebuilding the same infrastructure: camera integration, control loops, teleoperation with GELLO or Quest controllers, data collection pipelines. Anvil's stack integrates with ROS2, FoxGlove, and LeRobot out of the box. The data collection and model training workflow works end to end, the company says, without writing new code.
The Physical AI framing is worth pausing on. The term has become a catch-all for the intersection of large neural networks and robot bodies, the idea that the same scaling logic that produced capable language models could produce capable manipulators and walkers if you have enough data from real physical interactions. Companies like Physical Intelligence, Figure, and 1X have raised heavily on this thesis. Anvil is not building foundation models. It is building the workbenches that other companies use to generate the data those models need.
This is a meaningful distinction. Devkit companies live and die on whether the hardware is actually good enough for serious research, not just impressive at demos. Anvil's approach, leaning on established open-source arm designs like OpenARM and keeping the stack modular, reduces the risk that a customer gets locked into a platform that stalls. It also means Anvil's revenue is tied to the growth of a developer ecosystem rather than a single robot product roadmap.
The robotics devkit space is not empty. Reach Robotics, Rainbow Robotics, and a handful of academic spinouts sell similar configurations. But the $5.5 million round appears to be one of the smaller recent bets in a category where Figure raised $1 billion in a single round and 1X closed a $125 million Series B. Anvil is operating at a different scale, targeting the individual team or lab that cannot afford a custom integration shop but also cannot wait for a $100,000 humanoid platform to ship.
Xia and Pradeep are clearly targeting a specific buyer: the robotics PhD student or early-stage startup that is building something genuinely new and does not want to spend their first six months soldering. That buyer is real and underserved. Whether Anvil can scale beyond them and become the standard infrastructure layer for Physical AI teams will determine whether this is a niche business or a category-defining company.
The thing to watch next is whether Anvil expands beyond arms. The current product line covers manipulators and sensor modules, but no mobile platforms or whole-body systems. If the company moves up the stack, it will be competing directly with the robot OEMs it currently counts as customers.
Sources: GNews AI Funding | Anvil Robotics | Crunchbase
† Add † footnote: "Source-reported; not independently verified."
†† Add † footnote: "Source-reported; not independently verified."
††† Add † footnote: "Source-reported; not independently verified."
†††† Add † footnote: "Source-reported; not independently verified."
††††† Add † footnote: "Source-reported; not independently verified."
† Add † footnote: "Source-reported; not independently verified."
†† Add † footnote: "Source-reported; not independently verified."
††† Add † footnote: "Source-reported; not independently verified."
†††† Add † footnote: "Source-reported; not independently verified."
††††† Add † footnote: "Source-reported; not independently verified."
