--- title: Gemini Robotics ER 2: what's new and how it works description: Google DeepMind released Gemini Robotics ER 2, an embodied reasoning model designed as the planning brain for robots, built on Gemini 3.5 Flash. Key... canonical: https://daily.dev/posts/gemini-robotics-er-2-what-s-new-and-how-it-works-fn0jvui7o twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Gemini Robotics ER 2: what's new and how it works | daily.dev og:description: Google DeepMind released Gemini Robotics ER 2, an embodied reasoning model designed as the planning brain for robots, built on Gemini 3.5 Flash. Key... og:url: https://daily.dev/posts/gemini-robotics-er-2-what-s-new-and-how-it-works-fn0jvui7o og:image: https://api.daily.dev/og/posts/fn0JVUI7O.png og:image:alt: Gemini Robotics ER 2: what's new and how it works og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Gemini Robotics ER 2: what's new and how it works **[Collections](https://daily.dev/sources/collections)** · 3 min read · 3 upvotes · 0 comments ## Summary Google DeepMind released Gemini Robotics ER 2, an embodied reasoning model designed as the planning brain for robots, built on Gemini 3.5 Flash. Key improvements over ER 1.6 include continuous video understanding (57.4% accuracy on five-stage task tracking), precision moment-finding at 91.3% accuracy with 4x faster execution, and multi-robot coordination without custom integration layers. The model reasons and plans concurrently with action execution rather than pausing between steps, enabling smoother robot behavior as demonstrated in a Boston Dynamics Spot demo. It connects to the Gemini Live API for sub-second bidirectional streaming, supports tool orchestration including Google Search and user-defined functions, and can hand off motor execution to any VLA model. Available via the Gemini API, Google AI Studio, and the Gemini Enterprise Agent Platform in two variants. ## Content Google DeepMind has released Gemini Robotics 2, a suite of three AI models aimed at giving robots more capable, general-purpose control. Here's what's actually in it. ## The three models **Gemini Robotics ER 2** is the embodied reasoning layer — the high-level brain. It handles multi-step task planning, communicates with humans, monitors its own progress, and coordinates multiple robots working together. It's built on Gemini 3.5 Flash and available now to developers via the Gemini API, Google AI Studio, and the Gemini Enterprise Agent Platform. Two variants exist: `gemini-robotics-er-2-preview` and `gemini-robotics-er-2-streaming-preview`. **The flagship VLA (vision-language-action) model** drives a full humanoid body — legs, torso, arms, hands — from a single natural language prompt. This is a meaningful step up from the previous version, which only handled upper-body control. It's currently available through a trusted tester program. **An on-device variant** runs locally without internet connectivity. It can adapt to a new robot body with fewer than 200 examples in a few hours, which matters for reliability and safety in environments where cloud connectivity isn't guaranteed. ## What ER 2 actually does better Compared to ER 1.6, the main improvements are: - **Continuous video understanding**: the model now operates on raw video feeds rather than static snapshots, so it can catch mid-execution failures — spills, slips, misalignments — as they happen. It hits 57.4% accuracy tracking five-stage task progress on video. - **Precision moment-finding**: 91.3% accuracy identifying critical task transitions, at 4x faster execution speed than before. - **Multi-robot collaboration**: different machines can coordinate through shared semantic understanding, with the model orchestrating handoffs between them. - **Tool use**: supports Google Search and user-defined functions, and can hand off motor execution to any VLA model. - **Live API integration**: bidirectional streaming with sub-second latency, which is why the Boston Dynamics Spot demo looks less jerky than previous robot demos — no awkward freezes while the model catches up. - **Instrument reading**: now handles digital displays, linear scales, rulers, and liquid thermometers, not just circular dials. - **Safety**: better adherence to physical constraints and improved human proximity detection. Sample code and the Spot demo are on GitHub. ## What's still genuinely hard The demos are impressive, but the team is candid about the gaps. Dexterous manipulation remains unsolved. Tasks like tying a trash bag succeed roughly 40–44% of the time. The core problems are contact-rich physics that are difficult to simulate accurately, and the sheer complexity of hands — 22+ degrees of freedom each. There's also a data problem. High-quality teleoperation data is expensive and slow to collect. Human video is abundant but doesn't translate cleanly to robot actuation. This "embodiment gap" is one of the main reasons progress is slower than the demos suggest. DeepMind's own researchers estimate general-purpose robots in daily life are 5–10 years away, with dexterous manipulation as the primary bottleneck. Manufacturing scale-up is the next challenge after that. ## Safety and geopolitics DeepMind also released ASIMOV-Agentic, a new benchmark for measuring agentic safety in robotic systems. One complication worth noting: many of the robot bodies compatible with these models are manufactured in China, which the US government is moving to restrict on security grounds. That's a supply chain problem that no amount of model improvement resolves. ## What to watch The streaming API is probably the most immediately interesting thing for developers — low-latency bidirectional communication opens up real-time human-robot interaction that wasn't practical before. The multi-robot orchestration capability is also new territory; coordinating a team of different machines through a shared reasoning layer is a different kind of problem than controlling a single robot well. --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#computer-vision](https://daily.dev/tags/computer-vision), [#robotics](https://daily.dev/tags/robotics), [#google-gemini](https://daily.dev/tags/google-gemini) [View this post on daily.dev](https://daily.dev/posts/gemini-robotics-er-2-what-s-new-and-how-it-works-fn0jvui7o) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Gemini Robotics ER 2: what's new and how it works","url":"https://daily.dev/posts/gemini-robotics-er-2-what-s-new-and-how-it-works-fn0jvui7o","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/gemini-robotics-er-2-what-s-new-and-how-it-works-fn0jvui7o"},"datePublished":"2026-07-30T15:27:35.260Z","dateModified":"2026-08-01T20:18:20.103Z","description":"Google DeepMind released Gemini Robotics ER 2, an embodied reasoning model designed as the planning brain for robots, built on Gemini 3.5 Flash. Key...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/bb028f62dc2ba8a9afaf66ba59dd9d69?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/bb028f62dc2ba8a9afaf66ba59dd9d69?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/gemini-robotics-er-2-what-s-new-and-how-it-works-fn0jvui7o","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,computer-vision,robotics,google-gemini","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Gemini Robotics ER 2: what's new and how it works"}]} ```