Gemini Robotics
In plain wordsA family of AI models from Google DeepMind, built on its Gemini models, that let robots understand what they see and what people ask of them, and turn that understanding into physical actions.
A family of robot foundation modelsrobot foundation model, n.: One large AI model trained on huge amounts of robot and internet data so that it can control many robots on many tasks, and be adapted quickly to new ones. from Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold. that extend the multimodal GeminiGemini, prop. n.: Google's family of AI models and its chatbot assistant, launched in December 2023. Gemini models handle text, images, audio and video and power many Google products. models to physical action. The line pairs two kinds of model. A vision-language-action modelvision-language-action model, n.: An AI model that looks through a robot's cameras, reads an instruction such as 'put the cup in the sink', and turns both straight into the robot's movements. (Gemini Robotics proper) maps camera images and natural-language instructions directly to motor commands, aiming at generality across objects, scenes and instructions, at dexterity, and at adaptation to new robot bodies. An embodied-reasoning model (Gemini Robotics-ER) is a vision-language modelvision-language model, n.: An AI model that understands images and text together — it can describe a photo, answer questions about a chart or find the picture that matches a sentence. specialized in spatial understanding, such as pointing, 3D object detection, grasp and trajectory prediction, task planning and success detection, which can orchestrate robot skills and call tools as an agent. Demonstrations began on bi-arm ALOHA 2 research platforms and extended to industrial arms and humanoids. Later releases added a version that runs locally on the robot, reasoning before acting with skill transfer between robot bodies (1.5, September 2025), and whole-body humanoid control and collaboration between different robot types (2, July 2026).
We use Gemini Robotics-ER to plan the sorting task and our own controller to execute each grasp.
Origin
Announced by Google DeepMind on 12 March 2025, with the technical report 'Gemini Robotics: Bringing AI into the Physical World' and a partnership with the humanoid maker Apptronik. An on-device variant followed in June 2025, the 1.5 models in September 2025, and Gemini Robotics 2, ER 2 and On-Device 2 at the end of July 2026.
Usage note
Access has been staged: the embodied-reasoning models were offered to developers through Google's AI platforms, while the action models went first to trusted testers and early-access partners. The 2025 launch also introduced ASIMOV, a benchmark for judging whether robot actions are semantically safe.
Related
- vision-language-action modelvision-language-action model, n.: An AI model that looks through a robot's cameras, reads an instruction such as 'put the cup in the sink', and turns both straight into the robot's movements.
- robot foundation modelrobot foundation model, n.: One large AI model trained on huge amounts of robot and internet data so that it can control many robots on many tasks, and be adapted quickly to new ones.
- GeminiGemini, prop. n.: Google's family of AI models and its chatbot assistant, launched in December 2023. Gemini models handle text, images, audio and video and power many Google products.
- Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold.
- π0π0, prop. n.: A general-purpose AI model for controlling robots, released in 2024 by the start-up Physical Intelligence, that sees through cameras, follows typed or spoken instructions and moves robot arms deftly enough to fold laundry.
- humanoid robothumanoid robot, n.: A robot built roughly in human shape, with a head, torso, two arms and usually two legs, so that it can work in spaces and with tools designed for people.
- embodied AIembodied AI, n.: AI that has a body, usually a robot and sometimes a simulated character, and learns or acts by sensing and moving in a world rather than only processing text or images.
- RT-2RT-2, prop. n.: A 2023 Google DeepMind model that controlled robot arms with a chatbot-style vision-and-language model, letting robots follow instructions about objects and ideas they had never practiced with.
- π0π0, prop. n.: A general-purpose AI model for controlling robots, released in 2024 by the start-up Physical Intelligence, that sees through cameras, follows typed or spoken instructions and moves robot arms deftly enough to fold laundry.
Sources
- Gemini Robotics Team, Google DeepMind (2025). Gemini Robotics: Bringing AI into the Physical World. arXiv preprint.
Mentioned in
π0π0, prop. n.: A general-purpose AI model for controlling robots, released in 2024 by the start-up Physical Intelligence, that sees through cameras, follows typed or spoken instructions and moves robot arms deftly enough to fold laundry.