Google DeepMind's Gemini Robotics ER 2 coordinates robot tasks by combining incoming observations with planning and tool calls. The launch article describes a high-level model that delegates physical movement to lower-level action models or robot-control interfaces.
Continuous video gives the coordinator information about how an action is progressing, allowing it to decide whether to continue, retry or move to another step. Developers expose movement capabilities as callable tools and stream observations into the model. This division lets a planning system coordinate existing robot controls while using visual feedback to assess the ongoing task.
Terms in this piece · Glossary
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
