Google launched Gemini Robotics ER 2 on July 30, making the embodied-reasoning model available through the Gemini API and Google AI Studio. Google reports 57.4% accuracy on progress classification and 91.3% on moment finding, but independent benchmark results were not available in the retrieved coverage. The model coordinates lower-level robot controls, monitors video for task progress and supports collaboration between robots.
Google launched Gemini Robotics ER 2 on July 30 as a high-level reasoning and orchestration model for robots. It is publicly available through the Gemini API and Google AI Studio, with a private preview on the Gemini Enterprise Agent Platform.
Unlike a vision-language-action model that directly produces motor commands, Gemini Robotics ER 2 is designed to plan and supervise multi-step work. Developers expose lower-level controls, navigation APIs or VLA models as tools, while streaming video, audio or text into the model. The system can reason about the next step while a robot is still acting, according to Google.
Tracking progress from live video
Google says ER 2 watches continuous video to estimate task progress, identify the moment a critical event occurs and decide when to move to the next step. Its examples include detecting whether a light bulb has been tightened, whether a trash bag has been tied and when to stop pouring into a cup.
In Google's evaluations, the model reached 57.4% accuracy when assigning video frames to five progress bands. It also reached 91.3% accuracy on moment finding, with a reported mean absolute timing distance of 0.96 seconds. These are company-reported results; the retrieved independent coverage confirmed the launch and product framing but did not provide a separate reproduction of the benchmarks.
Orchestration and multi-robot work
ER 2 can call tools such as robot-control interfaces, navigation APIs and user-defined functions. Google demonstrated the model orchestrating Boston Dynamics Spot APIs to fetch an object from a natural-language request. The company also showed an Apptronik Apollo 2 robot and a Franka F3 Duo collaborating in a shared workflow.
Axios describes the release as part of Google's broader Gemini Robotics 2 update, which includes both lower-level action models and systems for higher-level planning. That distinction matters for engineering teams: ER 2 coordinates and monitors work, while a separate controller still performs the physical motion.
Google also reports gains on safety-instruction following and human-proximity tests, including stopping a humanoid when a person enters the work area and resuming after the area clears. Those results are useful product evidence, but deployments still require robot-specific safety controls, validation and human oversight rather than treating a foundation-model benchmark as a complete safety case.
Key Points #
- 1Gemini Robotics ER 2 is publicly available through the Gemini API and Google AI Studio as a high-level orchestrator for robot controls and VLA models.
- 2Google reports 57.4% progress-classification accuracy and 91.3% moment-finding accuracy; the retrieved independent coverage did not reproduce those benchmarks.
- 3The release adds video-based progress tracking and multi-robot collaboration, while physical deployments still require robot-specific safety validation.
Scoring Rationale #
Gemini Robotics ER 2 is a developer-accessible model release that adds real-time video progress tracking, tool orchestration and multi-robot coordination. It is materially relevant to robotics and embodied-AI teams, while benchmark and safety claims remain vendor-reported.
Sources #
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.