AWS Unveils 'Physical AI Toolchain' for Intelligent Machine Development

On the 12th, AWS unveiled the open-source stack "AWS Physical AI Toolchain on AWS" to support the development of intelligent machines. The stack is aimed at Physical AI fields such as robotics and autonomous systems. Physical AI refers to technology that allows machines to perceive the real world, make decisions, and act directly.

The toolchain is built by combining AWS services with NVIDIA Physical AI software. Its components include architecture guidance for Physical AI development, deployment automation, and reference code. It supports the entire process, from data generation and model training to simulation and edge deployment.

AWS said that depending on the deployment environment, such as production lines, warehouses, and robots, the data, models, and hardware required can vary, making the development process more complex. It said it created the toolchain as a common development foundation to ease that complexity.

The toolchain consists of five areas: synthetic data generation, model training, simulation and validation, edge deployment, and continuous improvement. It first operates by generating training data in a virtual environment and training models based on that generated data. It then proceeds to verify simulated operation before applying the system to actual equipment.

After that, it collects operational data from machines deployed in the field and uses the collected operational data for retraining. Through this structure, it repeatedly improves models by reflecting data accumulated in real-world environments.

Throughout this process, key technologies from AWS and NVIDIA are used together. AWS technologies include Amazon SageMaker, EC2 GPU instances, AWS IoT Greengrass, and Amazon Bedrock AgentCore. NVIDIA technologies include Isaac Sim, Isaac Lab, GR00T, and Cosmos.

Companies can use a single development environment for the entire stack, or integrate only the functions they need into existing workflows. In addition, based on machine fleet management capabilities, wireless provisioning for equipment on the scale of several thousand units is possible, and security policy management and software update management are also possible.

Based on experience operating more than 1 million robots in Amazon's logistics network, AWS designed the toolchain to reflect its own robotics operations experience. AWS incorporated the experience it accumulated in robot operations into the architecture guidance and deployment methods.

The toolchain was developed under a plan to reduce the time and burden of infrastructure setup for manufacturers and robotics companies. AWS plans to help manufacturers and robotics companies focus on model and product development. Uwem Ukpong, vice president of AWS Industry, said Physical AI will affect the entire goods movement, manufacturing, and production industries. He added that the company will help customers devote more of their capabilities to real innovation and product development instead of infrastructure setup.

NVIDIA said that linking training, simulation, and deployment platforms is important in Physical AI. Amit Goel, head of NVIDIA's robotics developer ecosystem and edge AI products, emphasized the importance of this linkage.

Amit Goel, head of NVIDIA's robotics developer ecosystem and edge AI products, said AWS Physical AI Toolchain combines AWS services with NVIDIA models, tools, and libraries. He said AWS Physical AI Toolchain supports the development and deployment of robotics applications.

Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=408038

References

• https://www.epnc.co.kr/news/articleView.html?idxno=408038