Google and Boston Dynamics Integrate Gemini AI Models Into Spot Robot for Natural Language Control and Task Execution

Gate News message, April 16 — Google and Boston Dynamics announced an integration of Gemini Robotics embodied reasoning models into the Spot quadruped robot, enabling the system to interpret natural language instructions, recognize objects, and execute complex tasks autonomously.

The collaboration, building on experimental work from a 2025 internal hackathon, allows Spot to process visual input and translate high-level commands into coordinated physical actions. Unlike traditional robotics programming, which relies on rigid step-by-step logic, the system uses conversational prompts and an intermediary software layer connecting Gemini models to Spot’s API. The AI can select from predefined capabilities including navigation, object detection, image capture, grasping, and placement, with real-time feedback enabling adaptive behavior without manual intervention.

In demonstrations, the system successfully interpreted general instructions to organize items, analyzing visual data and directing the robot through action sequences. The approach maintains operational safety by restricting AI capabilities to predefined boundaries within the robot’s API, balancing adaptability with controlled performance. The partnership suggests potential efficiency gains for developers by reducing manual coding and allowing engineers to focus on defining objectives rather than programming every action sequence.