AWS launches open-source Physical AI Toolchain

The launch addresses a practical gap in Physical AI: unlike large language models that can train on vast internet text, robots often lack enough real-world data. AWS's Sri Elaprolu noted that generating synthetic data representative of the real world is one of the problems the toolchain aims to solve, along with simulating and validating models before they reach physical hardware. The offering also draws on lessons from Amazon's own operations, which now include more than 1 million robots, though Elaprolu pointed out that controlled fulfillment centers don't represent every environment. He cited Telexistence, a Japanese robotics company deploying humanoids in convenience stores, with more than 300 already deployed, as an example where fault tolerance is lower and response rates must be higher. The toolchain stays hardware-neutral, which may let developers adapt it to specialized needs such as tactile sensing, a field where AWS worked with South Korean company RLWRLD to build a model for five-finger dexterity tasks. It also supports autonomous construction robots from Bedrock Robotics, whose models are trained in the cloud on AWS.