Humanoid robots have just begun stepping into Amazon warehouses and Mercedes-Benz automotive factories. Now, they are being recruited for an even more ambitious effort 鈥� the creation of artificial general intelligence with capabilities comparable to those of humans.
US computing firm Nvidia, which has become one of the world鈥檚 most valuable companies through its AI chip sales, recently announced several hardware and software products to boost humanoid robot training. The centrepiece is a 鈥渕oonshot鈥� initiative, called , focused on developing an AI model to help humanoid robots learn more efficiently through language instructions, watching human demonstrations and practising various actions with the help of human teleoperators.
鈥淚鈥檓 very excited to work with multiple leading humanoid robot companies across the world so that GR00T may transfer across embodiments,鈥� said , a research scientist at Nvidia heading Project GR00T, during the company鈥檚 AI conference held in San Jose, California, this week.
Advertisement
Robotic interactions with the physical world could, in turn, improve the AI model through an 鈥渆mbodied intelligence鈥� strategy 鈥� a cornerstone of Nvidia鈥檚 road map for developing artificial general intelligence. 鈥淲hat we truly want are AI agents as versatile as WALL-E, as diverse as all the robot forms or embodiments in Star Wars, and that work across infinite worlds, virtual or physical, as in Ready Player One,鈥� said Fan.
Even if it falls short of that lofty goal, Nvidia stands to potentially profit by becoming a key provider of robotics hardware and software, including computer simulations for training the robots鈥� virtually and a new Jetson Thor computer for robots that will enable them to run more powerful AI models. Meanwhile, humanoid robotics companies seem eager to access Nvidia鈥檚 massive computing resources and AI expertise.
Training large-scale AI algorithms on massive amounts of data requires specialised hardware, 鈥渁nd Nvidia knows how to build that鈥�, says at Agility Robotics, a humanoid robotics company in Oregon. 鈥淭hey are the de facto provider for basically all of the other players in this space.鈥�
The Jetson Thor computer could help humanoid robots process more sensor data to better perceive their surroundings, while ensuring they comply with safety regulations, says Velagapudi. A successful Project GR00T that empowers robots to learn from natural language communication and mimicking humans could also 鈥渟ignificantly speed up our ability to move into new markets and increase the capabilities of the robot鈥�, he says.
Agility has already begun moving humanoid robots out of the lab and into commercial use. Since late 2023, the company has been testing its robot, called Digit, in warehouses, where it picks up and moves empty containers. Agility has also started a trial run in a warehouse owned by the womanswear brand SPANX and operated by . 鈥淎gility鈥檚 perspective has always been that the fastest path to getting to generalised, multipurpose capability is to start by incrementally taking it one piece at a time,鈥� says Velagapudi.
Similarly, the Texas-based company Apptronik has signed an agreement with automaker to explore how its Apollo humanoid robots could deliver parts and perform inspections on the assembly line in automotive factories, with the goal of starting commercial operations by early 2025.
鈥淭he world is designed for human beings, so the robots that will be most broadly effective will be able to operate like human beings,鈥� says at Apptronik. 鈥淭housands of specialised robots may be able to accomplish a task or two with great efficiency today, but their potential additional utility is extremely limited.鈥�
However, when it comes to delivering precise and reliable robotic performance, AI-powered learning still falls short of handcrafted solutions from human engineers, said , founder of the robotics company Boston Dynamics and current head of The AI Institute in Massachusetts, during an Nvidia conference session. That is why humanoid robots have yet to prove themselves as skilled as humans on most tasks. 鈥淲hen there is concern that robots are going to take everybody鈥檚 jobs, I just wish we could take someone鈥檚 job,鈥� he said. 鈥淲e鈥檙e right at the front edge of doing almost any useful work at all.鈥�
Although limitations remain, there is an undeniable resurgence of enthusiasm for humanoid robots. Interest in robots has accompanied the recent AI boom based on large language models and other generative AI programs 鈥� and Nvidia seems intent on accelerating that trend.
鈥淧reviously, it was thought that it would take several decades for general-purpose humanoid robots to become a reality,鈥� says at Fourier Intelligence, a robotics company in Shanghai, China. 鈥淣ow, looking ahead, it is likely that in the next five to 10 years, humanoid robots with universal capabilities will emerge.鈥�
