AWS releases an open-source Physical AI Toolchain for building robots
Sorted AI12 OctAmazon launched the Physical AI Toolchain on AWS on 8 October 2026, which it describes as an open-source stack for building robots and other machines that sense and act in the real world. It bundles architecture guidance, deployment automation and ready-to-use code covering synthetic data generation, model training, simulation and continuous improvement, using AWS services such as SageMaker with NVIDIA's Isaac Sim, Isaac Lab, Isaac GR00T and Cosmos. Edge deployment to machines, via IoT Greengrass, is marked as planned, not yet available, in its GitHub repository.
Teams can use the whole stack or pick parts such as simulation or training. Amazon says it draws on its experience running more than 1 million robots, and that it can help manufacturers launch physical AI capabilities in weeks rather than years. NEURA Robotics says it helps it move faster from learning to deployment.
Why it matters: If you build or automate machines on factory floors or in warehouses, this is free reference code to start from instead of wiring tools together yourself. Expect GPU costs on AWS; the code is free, the compute is not.
Amazon ↗AI-assisted, reviewed by Sorted AI
▲ 0
0 comments