Creating AI models for humanoids seems to be the next big thing, and Nvidia has teased its new foundation model for humanoid robots that it’s calling Project GR00T.
Just like how an LLM can be used by a chatbot tool to give answers, write code, understand context, and do reasoning, a specially developed model is required for humanoids to perform the expected basic tasks. A robot that’s going to be employed for household chores or a factory’s assembly line doesn’t need to know about Nietzsche’s philosophy, Roman conquests, and Napoleon’s wars. It needs to know things that might sound simpler to humans but are actually more complex to program. For example, comparing objects, arranging things by color, or identifying a slippery surface. Machines, being dumb as they are, find this stuff to be remarkably more challenging than recalling philosophy or modern history. Just built differently, that’s all.
A model that’s supposed to run a humanoid robot can’t be a simple LLM like GPT-4 or Gemini Pro. Nvidia’s GR00T (short for Generalist Robot 00 Technology) is all about learning natural language and emulating movements. It can mimic and learn from human actions. Coordination, dexterity, and skills to adapt and interact with the real world are key to this model’s goal.
Nvidia is doing excellent work in the field of robotics. The Isaac foundation models for robotic arms and 3D vision, for example, are already some of the finest glimpses of intelligent robotics technology. With GR00T, Nvidia is right up there against Tesla Optimus, Boston Dynamics Atlas, Figure AI Figure 01, and so on. CNET was quick to compare GR00T against Optimus.
In the official announcement, Nvidia teased the bundle of technologies powering its leap into the world of AI robots – “Robot Training Simulator, Jetson Thor Robot Computer, Generative AI Foundation Models, and CUDA-Accelerated Perception and Manipulation Libraries.”
Whether you’re a company in need of specialized SoCs or GPUs that can run multimodal, audio-visual models to enable robots to interact with the world around them or someone who just needs a generative AI model built for the robots of the future, Nvidia has you covered (the latter solution being Project GR00T).
The full GTC March 2024 Keynote delivered by Jensen Huang is 2 hours long, but you can watch a couple of cute robots at the 1 hour, 55-minute mark:
If you’re worried that robots are going to replace humans in many jobs, don’t be. They’ve already begun to do so.
Companies like Nvidia, Tesla, and a bunch of other engineering-focused enterprises are eyeing this new segment. In a bid to cut costs, everyone is already using generative AI to replace customer service, researchers, and writers using basic generative AI tools. This is just the physical extension of this trend – AI robots replacing menial, rote tasks that might be high-risk and don’t require much thinking.
As one would expect, the stock price of Nvidia shot up. The news was released on Monday, March 18 (the weekend being non-market days) and the company registered a handsome growth in the next 5 days.
