Watch the video
Whole-body control clip from the official model page, played as published. · Google DeepMind · 1080p source video
Mechanism and experiments
Gemini Robotics uses a vision-language-action model to turn observations and instructions into robot actions. Its whole-body video shows container carrying and movement, alongside separate dexterity and multi-robot demonstrations. Gemini Robotics ER focuses on spatial reasoning and multi-step planning, with execution connected to a robot control system.
Test conditions and scope
Model demonstrations, ER API access and downloadable weights are different levels of disclosure. The official page supplies model descriptions, recordings and access routes; weights absent from that page are not labelled open source. The demonstrated task is distinct from constrained-space servicing.