Google announced that Aaron Saunders, former CTO of Boston Dynamics, has joined DeepMind as Vice President of Hardware Engineering, responsible for expanding Gemini into a cross-platform robot operating system. Saunders previously led Atlas backflips and Spot mass production. After joining, the Gemini Robotics team will accelerate the hardware implementation of the "plug-and-play" general AI system.

image.png

Gemini Robotics Roadmap  

- June 2025 Gemini Robotics: Locally optimized version, supporting real-time multimodal reasoning  

- September 2025 Gemini Robotics 1.5: Introduce "embodied reasoning," end-to-end visual-language-action  

- February 2026 Gemini Robotics 2.0: Led by Saunders, release a unified hardware abstraction layer (HAL), open APIs and SDKs to third-party robot manufacturers

Hardware Strategy  

Saunders revealed that DeepMind is developing the "Gemini Control Hub"—a compact computing box integrating SoC + NPU + ISP, supporting USB-C plug-and-play, aiming to reduce inference latency to 50ms, power consumption less than 15W, and can be embedded in various forms such as humanoids, robotic arms, and drones.

Ecosystem Goals  

Hassabis said, "We hope Gemini becomes the Android of the robotics industry—same AI, adapted to all bodies." Google plans to launch a hardware certification program in Q2 2026, robots that pass the certification can be pre-installed with Gemini Robotics system and get an app distribution channel similar to the Play Store.

Market Impact  

Analysts point out that if the Gemini Robotics HAL is released as planned, robot manufacturers do not need to develop their own AI stacks, and can enjoy multimodal perception, task planning, and cloud update capabilities, which could reduce software R&D costs by 50% and accelerate the commercialization of humanoid robots.