Gemini Robotics 2 by Google DeepMind

Gemini Robotics 2 is a family of AI models from Google DeepMind that act as the intelligence layer for advanced robotics. It comprises three variants: a full vision-language-action model for whole-body and dexterous manipulation, an embodied reasoning model for multi-step task planning, and an efficient on-device model that adapts to new robot bodies in hours. This is not a standalone startup but a deep-tech research project within Google, aimed at solving the critical challenge of creating general-purpose, adaptable robot intelligence. Its commercial path will likely be via cloud APIs and enterprise partnerships, leveraging Google's ecosystem. For a founder analyzing this, the key takeaway is that the barrier to entry in this space is extremely high, but niche opportunities exist in vertical-specific robot applications built on such foundational models once they become accessible.