Physical AI reshaping manufacturing robotics
Physical AI—robotic systems using Robotic Foundation Models to adapt and learn on the job—is being touted as manufacturing’s next advantage, promising less reprogramming and more on-the-fly autonomy for factory robots argued. That trend shifts verification and control design toward sim-to-real transfer, HIL testing, and AI-driven motion planning.
On Jan 5, 2026 NVIDIA released new open physical‑AI models—Cosmos Transfer 2.5, Cosmos Predict 2.5, Cosmos Reason 2—and the Isaac GR00T N1.6 model, and published them on Hugging Face for robotics development. nvidianews.nvidia.com Microsoft announced Rho‑alpha on Jan 21, 2026, a Phi‑series derived robotics model for bimanual manipulation being offered through a Research Early Access Program. news.microsoft.com Microsoft also published a Physical AI toolchain on GitHub and has publicized partnerships with Hexagon and other firms to deploy Azure‑backed physical‑AI solutions in manufacturing. microsoft.github.io news.europawire.eu Hardware‑in‑the‑Loop is moving into mainstream robotics pipelines: NVIDIA’s Isaac Sim includes HIL fundamentals and ROS 2 integration modules for testing real sensors and actuators against simulated environments. docs.nvidia.com Siemens’ Simcenter Testlab RT explicitly advertises real‑time physical‑virtual testing for mechatronic subsystems, and MathWorks documents HIL as a standard step for embedded controller validation. siemens.com mathworks.com NVIDIA also shipped an updated Jetson T4000 module and an OSMO edge‑to‑cloud training workflow built on the Blackwell architecture, which the company says improves energy efficiency and on‑device AI throughput. nvidianews.nvidia.com Franka Robotics, NEURA and Humanoid are cited as early adopters using GR00T‑enabled workflows to simulate, train and validate new robot behaviors. nvidianews.nvidia.com
