Meta has acquired Assured Robot Intelligence, a San Diego-based startup building foundation models for humanoid robots, folding the team into its Superintelligence Labs research division as the company accelerates its push into physical artificial intelligence.
The deal, announced on Friday for an undisclosed sum, brings co-founders Lerrel Pinto and Xiaolong Wang into Meta's internal robotics effort, where they will work alongside the Meta Robotics Studio team established last year to develop foundational technologies for humanoid systems.
ARI was building AI systems that enable robots to understand, predict and adapt to human behaviour in complex environments, using learning-based control systems rather than rigid pre-programmed instructions.
The startup had raised a seed round from AI specialist firm AIX Ventures and was developing whole-body robot control models alongside e-Flesh, a tactile sensor that measures deformations in 3D-printable microstructures using magnets, addressing one of the unsolved problems in making robots interact safely with physical objects.
Both founders bring significant pedigrees.
Pinto previously taught at New York University and co-founded Fauna Robotics, a humanoid startup Amazon acquired in March to bolster its own robotics effort.
Wang is an associate professor at UC San Diego, a former Nvidia researcher and winner of the MLSys 2024 Best Paper Award.
Meta's strategy differs from competitors such as Tesla, which is vertically integrating design, manufacturing and sales of its Optimus humanoid.
Instead, Meta intends to develop sensors, software and AI models and make them available to the broader industry, functioning as a platform provider rather than a hardware manufacturer, a model comparable to Google's Android in smartphones.
The company would supply the intelligence layer and capture value through data, the model ecosystem and integration with its existing platforms, where 3.3 billion people interact daily.
Alexandr Wang, head of Meta Superintelligence Labs, said the team aims to achieve "physical AGI," adding that scaling "will come from learning directly from human experience, not teleoperation alone."
"We have the potential to transform AI that can think and talk to AI that can do, assisting humans safely and reliably in the physical world," Pinto wrote on X.
Meta's MIA500 machine learning chip, unveiled last month, may also factor into its robotics ambitions.
The inference accelerator delivers 10 petaflops of FP8 performance and could theoretically be optimised for deployment inside humanoid hardware built by third parties.
The acquisition reflects an intensifying race across the sector.
Tesla plans to begin large-scale production of its Optimus V3 humanoid between July and August, targeting annual capacity of one million units at a price between $20,000 and $30,000.
1X Technologies has opened a factory in California to produce 10,000 NEO robots in its first year, with first-year capacity selling out within five days of preorders.
Figure AI, backed by Nvidia, OpenAI and Jeff Bezos, is targeting 100,000 deployments over four years, and Apptronik has raised $520 million at a $5 billion valuation in partnership with Google DeepMind.
Goldman Sachs projects the humanoid market will reach $38 billion by 2035, while Morgan Stanley estimates it could grow to $5 trillion by 2050.
The recap
- Meta acquired Assured Robot Intelligence; financial terms undisclosed.
- Morgan Stanley projects humanoid market at $5 trillion by 2050.
- Meta aims for "physical AGI" learning from human experience.
