Join waitlist
Back to products
Gemini Robotics

Gemini Robotics

Bringing AI into the Physical World

Website deepmind.google
Overview

What it is

Gemini Robotics from Google Deepmind, is the Gemini 2.0-based AI models for robots. Multimodal, general, interactive, and dexterous. Powers ALOHA 2, Apptronik Apollo, and more.

Intent

I need it when

Perform dexterous manipulation tasks requiring fine motor control and precision (e.g., folding origami, packing, food preparation)

Gemini Robotics-ER 1.6 is specifically designed for embodied reasoning and dexterous skills, enabling robots to tackle complex tasks requiring fine motor coordination. The model learns to plan and execute precise manipulations across different robot morphologies.

Enable robots to perform complex multi-step physical tasks autonomously with minimal retraining across different robot types

Gemini Robotics provides vision-language-action (VLA) and embodied reasoning (ER) models that allow robots to perceive environments, reason about tasks, and execute motor commands. The models generalize across multiple robot embodiments (bi-arm, humanoid, etc.), enabling a single model to work on different platforms and solve novel tasks without task-specific retraining.

Reduce development time and cost for robotics applications by leveraging pre-trained foundation models instead of building custom solutions

Gemini Robotics offers pre-built models (Gemini Robotics 1.5 VLA, Gemini Robotics-ER 1.6, and on-device variants) that robotics developers can integrate via SDK or Google AI Studio. This eliminates the need to train models from scratch and allows teams to focus on application-specific adaptation rather than foundational model development.

Access early-stage AI robotics technology through a structured partnership or testing program with Google DeepMind

Gemini Robotics is available through a waitlist for early access and via the Google DeepMind Accelerator program, which supports early-stage robotics startups. Trusted testers and partners (Boston Dynamics, Agility Robotics, Universal Robots, etc.) can access models and provide feedback to guide product development.

Enable robots to understand and respond to natural language commands and adapt behavior in real-time based on environmental changes

Gemini Robotics supports multimodal reasoning (text, images, audio, video) and includes interactivity features that allow robots to understand everyday commands, explain their approach in natural language, and accept real-time user redirection without technical language barriers. This makes robots more intuitive and responsive to dynamic environments.

Drop

Not a fit when

  • When you need immediate commercial deployment—Gemini Robotics is in early access via waitlist and not yet generally available for production use
  • When your robots require specialized domain training that cannot be transferred across different robot embodiments
  • When you operate robots that are not compatible with the supported platforms (ALOHA, Bi-arm Franka, Apptronik Apollo, or other tested embodiments)
  • When you need on-premises-only deployment without cloud connectivity, as the models require integration with Google's infrastructure
  • When your use case involves safety-critical applications without human oversight, as the product emphasizes human redirection and transparency rather than full autonomy
Commercials

Pricing

Pricing not specified