Sou Hu Cai Jing·2026-02-06 21:19Core Insights - Microsoft Research has launched Rho-alpha, a new robotic model designed to help robots understand natural language commands and perform complex physical tasks in less structured environments [1] - Rho-alpha aims to advance the next generation of robotic systems, enabling them to perceive, reason, and act in dynamic real-world settings [1] - The model is part of a trend towards "visual-language-action" models that enhance the autonomy of physical systems [1] Group 1 - Rho-alpha integrates touch data and is currently being researched to support additional sensory modalities, such as force sensing [2] - The model is designed to improve continuously during deployment by learning from user feedback during interactions with robots [2] - The training of Rho-alpha heavily relies on synthetic data, utilizing a multi-stage training process that combines reinforcement learning and simulation technology [2] Group 2 - A major challenge for foundational models is the lack of diverse real-world robotic data [4] - Researchers are collaborating with Microsoft to enhance pre-training datasets using synthetic demonstrations, addressing the impracticality of remote operation in many cases [4] - NVIDIA emphasizes the role of synthetic data in accelerating the development of robotic technologies, highlighting the collaboration with Microsoft to generate high-fidelity synthetic datasets [4] Group 3 - Microsoft has opened registration for the early access program for Rho-alpha and plans to release more updates on its robotic research in the coming months [4]