Meta has purchased Assured Robot Intelligence to develop the Android for humanoid robots.

      TL;DR: Meta has acquired Assured Robot Intelligence, a startup co-founded by Lerrel Pinto, a former Fauna Robotics co-founder, and Xiaolong Wang, a former Nvidia researcher, to enhance its humanoid robotics strategy. This acquisition integrates whole-body robot control models and tactile sensor technology into Meta Superintelligence Labs, reflecting Meta's goal to become the core intelligence provider for humanoid robotics, akin to Android's role for smartphones.

      Lerrel Pinto previously co-founded Fauna Robotics, which developed a user-friendly bipedal robot named Sprout, before leaving in 2025. Fauna was acquired by Amazon in March, along with its team and its compact dancing humanoid, as Amazon entered the consumer robotics sector. Pinto later teamed up with Xiaolong Wang, who has received recognition for his research on AI model optimization, to establish Assured Robot Intelligence. Meta announced the acquisition of ARI last Friday, with both founders joining its Superintelligence Labs. The financial details of the deal were not disclosed. While the location of ARI's employees in San Diego and New York is interesting, the focus is on Meta's intentions for the technology and its implications for the humanoid robotics market.

      The platform

      Meta aims to replicate the impact that Google's Android operating system and Qualcomm's chips had on the smartphone industry by building a foundational platform for robotics. The company launched Meta Robotics Studio last year, appointed former Cruise CEO Marc Whitten to lead the initiative, and began recruiting about 100 engineers to develop humanoid hardware and AI models in-house. The company’s CTO, Andrew Bosworth, has characterized humanoid robots as a significant venture for Meta, comparable to augmented reality, a sector in which Meta has already invested heavily through its Reality Labs division. The acquisition of ARI adds capabilities for robots to understand, predict, and respond to human behavior in chaotic settings.

      Meta’s strategy is clear: it plans to create sensors, software, and AI models for robots while making these resources accessible to external manufacturers. This approach mirrors the Android model as applied to physical machines. Just as Google offered its operating system for free while benefiting from search, advertising, and the Play Store ecosystem in smartphones, Meta aims to provide the intelligence layer for robotics and derive value from data, the ecosystem of models, and integration with its existing platforms that already have billions of daily users. The company has aggressively pursued AI talent, recently hiring five founding members from Thinking Machines Lab, with one researcher reportedly receiving a compensation package of $1.5 billion over six years. The ARI acquisition reflects this trend: a compact team with cutting-edge expertise that can be immediately integrated into the Superintelligence Labs.

      The technology

      ARI contributes technology focused on “robotic intelligence that enables robots to understand, predict, and adapt to human behaviors in varied environments.” This involves AI models for controlling humanoids, allowing the robot to coordinate movements, maintain balance, and react based on real-time sensory input. Wang's recognized work on activation-aware weight quantization—an essential method for making AI models efficient enough to operate within a robot's limited computing capacity—plays a critical role here.

      The company also developed e-Flesh, a tactile sensor that utilizes magnets and magnetometers to gauge distortions in 3D-printed microstructures. Tactile sensing is one of the major challenges in humanoid robotics, as a robot reliant on cameras and lidar cannot differentiate between grasping an egg and a tennis ball without tactile feedback. The disconnect between simulated robot learning and real-world execution remains a significant barrier to widespread deployment. ARI’s advancements in self-learning for robot control and its sensory technology help bridge this gap with improved models and enhanced sensory input.

      The market

      The humanoid robotics field has transitioned from speculative to competitive within just 18 months. Tesla is set to start large-scale production of its Optimus V3 humanoid in mid-2026, aiming for an annual output of one million units at a price of $20,000 to $30,000. 1X Technologies has opened a facility in Hayward, California, targeting a first-year production of 10,000 NEO humanoid robots, which sold out within five days of preorders. Apptronik has secured $520 million at a $5 billion valuation, collaborating with Google DeepMind and its Gemini Robotics models. Amazon has made two acquisitions in robotics in one month, while Unitree aims for 20,000 humanoid deliveries by 2026. According to Morgan Stanley, the global humanoid robot market is projected to reach $38 billion by 2035 and $5 trillion by 2050.

      The competitive landscape is falling into three tiers: the top tier comprises vertically integrated manufacturers like Tesla and 1X that design, build, and sell complete robots; the second tier consists of platform providers that offer intelligence layers, operating systems, or essential components for various manufacturers; and the third tier includes component suppliers, such as chip and sensor