Microsoft Introduces Rho-alpha Robotics Model for Physical AI

Microsoft has introduced Rho-alpha, a robotics model designed to advance AI for physical systems, announced on its research blog. The model is derived from Microsoft’s Phi series of vision-language models and aims to enhance robotic autonomy in unstructured environments.

Rho-alpha is described as a vision-language-action-plus (VLA+) model. It converts natural language commands into control signals for robotic systems performing tasks that require two arms, such as plug insertion or object manipulation. The model incorporates tactile sensing for perception and is being tested on dual-arm setups and humanoid robots.

Training for Rho-alpha combines real-world physical demonstrations with synthetic data generated through reinforcement learning in simulation environments using NVIDIA Isaac Sim on Azure. This approach addresses the limited availability of large-scale robotics datasets, especially those involving tactile feedback.

Organizations interested in evaluating Rho-alpha can apply for Microsoft’s Research Early Access Program. The company plans to make the model available later through Microsoft Foundry and will publish a technical description in the coming months.