Soon, Meta will be more than just a social media company. According to a recent report, the company is establishing a new division within its Reality Labs unit to develop AI humanoid robots.
This move comes ahead of Apple’s anticipated entry into the humanoid robotic space, which analysts believe is still in the proof-of-concept stage and unlikely to commence mass production before 2028.
Meta chief technology officer Andrew Bosworth said in a memo that the new robotics product group will focus on “consumer humanoid robots to maximise Llama’s platform capabilities.” Llama is Meta’s series of AI foundation models that power various generative AI products across its platforms.
“We believe expanding our portfolio to invest in this field will only accrue value to Meta AI and our mixed and augmented reality programs,” Bosworth wrote.
Meta plans to develop its own humanoid robot hardware, initially targeting household tasks. The company is also working on AI, sensors, and software for robots manufactured by other companies and has started discussions with robotics firms, including Unitree Robotics and Figure AI. According to Bloomberg, Meta does not currently plan to launch a branded robot.
The division will be led by Marc Whitten, former CEO of self-driving car company Cruise, who has been hired as Meta’s vice president of robotics, according to the memo.
Meta has also hired John Koryl, former CEO of second-hand e-commerce company The RealReal, as vice president of retail. Koryl will focus on selling Meta’s Quest mixed reality headsets and AI wearables, such as Ray-Ban Meta smart glasses, through retail partnerships and direct-to-consumer channels.
Tech Giants Clash as Robotics Race Heats Up
Recently, analyst Ming-Chi Kuo noted that Apple’s entry into humanoid and non-humanoid robots was in the early stages and that it may take three more years to materialise.
“While the industry debates the merits of humanoid vs. non-humanoid designs, supply chain checks indicate Apple cares more about how users build perception with robots than their physical appearance (so Apple uses anthropomorphic instead of humanoid), implying sensing hardware and software serve as the core technologies,” said Kuo.
He added that Apple’s robot mass production will likely not start until 2028 or later. Interestingly, Apple has been unusually open about sharing some of its robotics research during the early POC stage, possibly to attract talent.
Similarly, NVIDIA is now turning its attention to robotics. The company recently launched Cosmos, a platform to accelerate the development of physical AI systems, such as autonomous vehicles (AVs) and robotics.
In a recent interview, NVIDIA chief Jensen Huang said that the world needs an AI that understands the physical world. “It has to understand the dynamics of the physical world, like gravity, inertia, or friction, and it has to understand spatial and geometric relationships,” Huang said.
On the other hand, Figure AI, backed by Microsoft, is highly bullish on humanoids. The company is shipping Figure-02 humanoid robots to its second customer and has built a new Robot/AI Campus. The company is also in talks to raise a massive $1.5 billion round at a $39.5 billion valuation.
The founder of the company, Brett Adcock, also recently took to X to say, “In our lifetime, you will see more humanoid robots than humans when you’re out and about.”
At the same time, Elon Musk also has big plans for Optimus. The bipedal robot has demonstrated advanced capabilities, such as walking on uneven terrain using neural networks. Elon Musk has described it as the ‘biggest product of any kind ever.’
Meanwhile, US-based humanoid robotics company Apptronik has raised $350 million in Series A funding to expand the production of its humanoid robot, Apollo, and meet growing customer demand.
As per the company’s blog, the funding round was co-led by B Capital and Capital Factory, and Google participated as well. This comes to light after the company’s latest partnership with Google DeepMind’s robotics team to create truly intelligent and autonomous robots.
Meta Research to aid Robotics
Meta recently introduced PARTNR, a research framework to improve collaboration between humans and robots. The goal is to make robots more intuitive and helpful in everyday tasks like cleaning, delivering items, and assisting with cooking.
The framework relies on large-scale simulation training before deploying robots in physical environments. Using Habitat 3.0, a simulator designed for human-robot interaction, robots learn to work alongside humans in a safe and scalable way. Meta has tested this model on Boston Dynamics’ Spot robot, demonstrating its ability to work alongside humans.
Last year, Meta FAIR introduced Meta Motivo, a behavioural foundation model for controlling virtual physics-based humanoid agents in complex whole-body tasks.
The model mimics human-like behaviours and performs as well as task-specific methods, exceeding state-of-the-art unsupervised reinforcement learning and model-based approaches.
Meta’s Fundamental AI Research (FAIR) also released three new research artefacts that advance touch perception, robot dexterity and human-robot interaction, namely Meta Sparsh, Meta Digit 360, and Meta Digit Plexus.
Besides this, Meta Reality Labs Research also released the Nymeria dataset publicly. The dataset provides egocentric human motion in the wild at an unprecedented scale, capturing a broad spectrum of people engaging in everyday activities across varied locations.
The Nymeria dataset records real-world human motion using a combination of multimodal egocentric devices, including Project Aria glasses and miniAria wristbands. This network of sensors simulates the types of data that future wearable technologies, such as AI-powered glasses and smartwatches, could leverage.
