Amazon Web Services (AWS) said on Oct. 12 it is introducing an open-source stack called the 'Physical AI Toolchain on AWS' to build intelligent machines.

The toolchain provides architecture guidance, deployment automation and ready-to-use code as open source, the company said. It integrates with Nvidia's physical AI stack to support the full physical AI development lifecycle.

It covers five core areas of physical AI development: synthetic data generation, model training, simulation and validation, edge deployment, and continuous improvement. It also enables operating data from machines in the field to be reused to generate new training data.

More companies are also developing physical AI on AWS. Neural Robotics is developing a cognitive humanoid robot that learns through seeing, hearing and experience, and aims to bring millions of intelligent robots to the market by 2030.

RealWorld is taking on dexterous manipulation, one of the challenges in physical AI. It is developing a foundation model with 8.1 billion parameters to help robot hands precisely grasp, turn and handle objects like humans in factories and service sites.

Config built a data pipeline that collects more than 200,000 hours of robot motion data. It uses generative AI to expand the collected data into various training scenarios, supporting stable operation even in real-world environments that are difficult for machines to predict.

Uwem Ukpong (우웸 우퐁), vice president for AWS industry, said, "Physical AI will impact every industry that moves, builds and makes something, and customers are moving quickly to seize this opportunity." He added, "We heard that customers are putting too much engineering capacity into building infrastructure rather than innovation, so we developed the AWS Physical AI Toolchain. We want to flip that balance."