What is physical AI, the tech giant's bet to accelerate the creation of humanoid robots

Nvidia, the chip and semiconductor giant, is preparing for what comes after generative artificial intelligence, marked by the launch of services like ChatGPT, Gemini, and Copilot. The company’s new bet is on physical AI.

The idea is to enable humanoid robots and devices like robotic arms in factories to perform complex actions in the real world, explained Marcio Aguiar, Nvidia’s Executive Director for Latin America, this Monday (28) during Web Summit Rio.

In March, Nvidia launched Isaac Gr00t N1, a physical AI model that serves as a foundation for other companies to build their own humanoid robots.

According to the company, the innovation helps speed up the creation of such devices because the model is pre-trained to grasp, move, and pass objects from one hand to another, as well as carry out multi-step tasks.

Developers and researchers can then apply post-training to tailor the model to specific tasks within their context.

Nvidia also states that other generative AI models, such as OpenAI’s GPT or Meta’s LLaMA, do not understand the physical world as effectively, which limits the capabilities of robots trained on those platforms.

Reinforcement Learning

With the concept of physical AI, models are trained to better understand space and the physics of the real world.

This training uses a technique called reinforcement learning, which allows devices to detect unexpected situations and find solutions quickly.

This enables robots to become skilled at tasks such as packing boxes, assembling vehicles, or navigating environments autonomously.

Source: G1

 Like this content? Follow us on social media and stay updated!