Unlocking the Power of Gemini Robotics ER 2

RoboticsAI & Machine LearningAutomated Logistics

Imagine a world where robots can think, plan, and collaborate like humans. Welcome to the future of robotics, where Gemini Robotics ER 2 is revolutionizing the industry.

TL;DR
• Gemini Robotics ER 2 is a new model designed to act as a high-level brain for robots.
• It enables real-time spatial reasoning, multi-step task planning, and collaboration between different robots.
• Gemini Robotics ER 2 is now publicly available to developers via the Gemini API, Google AI Studio, and in private preview on Gemini Enterprise Agent Platform.

Summary
Gemini Robotics ER 2, released July 30 2026, delivers a high‑level reasoning engine for robots, enabling real‑time spatial awareness, multi‑step planning, and collaborative task execution. The model integrates with existing VLA systems and offers API access for rapid deployment today.
• Real‑time video tracking allows robots to self‑correct and advance tasks without pause.
• Multi‑robot collaboration enables diverse machines to share a semantic understanding and handoff work.
• Progress classification and moment‑finding achieve 57.4% and 91.3% accuracy, respectively, with sub‑second latency.
• Safety benchmarks show the model halts robots near humans and resumes only when the area clears.

Gemini Robotics ER 2 improves tool orchestration across real VLA, simulated VLA, and human tele‑op modes, outperforming ER 1.6 in all three. The Gemini Live API’s bidirectional streaming reduces latency, allowing fluid command execution and eliminating stop‑and‑think pauses. A Boston Dynamics Spot demo demonstrates the model orchestrating navigation and manipulation APIs to fetch objects on natural language commands.

The model’s continuous progress classification assigns each video frame to five progress levels, providing situational awareness that lets robots adjust actions on the fly. Precision moment‑finding identifies critical event frames with 91.3% accuracy and a 0.96‑second mean absolute distance, enabling accurate task switching and success verification.

Spatial reasoning advances include raw‑video success/failure detection, generalized instrument reading across ten display types, and enhanced visual question answering. Safety instruction following and human proximity benchmarks confirm the model’s ability to enforce constraints and detect nearby humans. Future work targets more complex tasks to accelerate helpful robot development and support the robotics community.

As we continue to push the boundaries of robotics, we must prioritize safety, collaboration, and innovation. With Gemini Robotics ER 2, we're one step closer to creating a future where robots and humans work together in harmony.