Boston Dynamics demos Gemini in Spot

Robots usually follow fixed scripts: map a route, stop at preset points, and run the same checks every time. Boston Dynamics is now showing Spot doing a looser kind of work, using Google DeepMind’s Gemini Robotics-ER 1.5 to interpret a room, break a chore into steps, and act on it. (bostondynamics.com) In the demo Boston Dynamics posted this week, Spot moves through a home, picks up shoes and soda cans, and sorts items into places like a shoe cabinet and a laundry basket. The company said the project grew out of a 2025 internal hackathon and was built as a proof of what a language-guided robot could do beyond a standard mission. (bostondynamics.com) Gemini Robotics-ER 1.5 is a vision-language model, which means it takes images and text and turns them into plans rather than directly moving motors. Google DeepMind says the model is designed for physical-space reasoning, including spatial understanding, multi-step planning, progress estimation, and tool use. (deepmind.google) Boston Dynamics said its engineers put a software layer between Gemini and Spot’s application programming interface, then exposed a limited set of tools: navigate to locations, capture images, identify objects, grasp them, and place them. Instead of writing a rigid state machine for every step, the team said it interacted with the model in conversational language. (bostondynamics.com) That is a shift from how Spot is commonly used in the field. Boston Dynamics’ software stack already lets operators record repeatable autonomous routes with GraphNav and Autowalk, replay them across robots, and attach actions to waypoints along the map. (github.com) The new layer adds planning on top of that motion system. Google DeepMind describes Gemini Robotics-ER 1.5 as a “high-level brain” that can turn a broad instruction like “clean the kitchen” into smaller steps, while another model or robot controller handles the physical motions. (deepmind.google) Spot can physically do this kind of work because the arm version already supports grasping, lifting, carrying, placing, dragging, door opening, and switch flipping. Boston Dynamics says the arm has six degrees of freedom, nearly one meter of reach, and a gripper camera, giving the robot both mobility and a way to act on what it sees. (bostondynamics.com) Boston Dynamics framed the home cleanup video as a demo, not a product launch. IEEE Spectrum reported that the company’s commercial focus for the Gemini partnership is industrial inspection, where Spot is already deployed and where Boston Dynamics says several thousand of its legged robots are working. (spectrum.ieee.org) That inspection push has already moved past the 1.5 demo model shown in the house. Boston Dynamics and Google DeepMind said on April 15, 2026 that Spot is being equipped with Gemini Robotics-ER 1.6 for tasks like reading gauges, spotting spills and debris, and reasoning about what it sees in industrial sites. (spectrum.ieee.org, therobotreport.com) The living-room cleanup clip is the easier way to see the idea: a quadruped robot that no longer needs every move spelled out in code. Boston Dynamics’ own description was simpler than that—Spot, with Gemini, was using language, images, and a short tool list to turn a to-do list into actions. (bostondynamics.com)