Google DeepMind Releases Gemini Robotics ER 2 for Advanced Embodied AI
Google DeepMind released Gemini Robotics ER 2 on July 30, 2026, a new embodied reasoning model that acts as a high-level brain for robots, enabling advanced capabilities like multi-step task planning, real-time progress tracking, and multi-robot collaboration.
Introducing Gemini Robotics ER 2
Gemini Robotics ER 2 represents Google DeepMind's most capable embodied reasoning model for robotics to date. It processes continuous video feeds, allowing robots to understand their environment and execute complex sequences of actions. This model moves beyond single-robot, single-task operations, facilitating collaborative workflows where multiple robots can work together and self-correct in real time.
Key Capabilities and Features
The ER 2 model suite is comprised of three distinct components: a vision-language-action model, the core embodied reasoning model (ER 2), and an on-device variant. This architecture supports a range of advanced functionalities:
- Multi-step Task Planning: Robots can break down complex objectives into a series of manageable steps.
- Real-time Progress Tracking: The system continuously classifies task progress from 0-100% in 20% increments, allowing for dynamic adjustments.
- Moment-Finding: This feature aids in task verification, ensuring that specific actions or milestones are achieved.
- Multi-robot Collaboration: ER 2 enables multiple robots to work together on tasks that a single robot cannot complete independently.
Integration with the Gemini Live API facilitates low-latency bidirectional streaming, crucial for responsive real-world robotic applications.
Performance Improvements Over Predecessors
Gemini Robotics ER 2 demonstrates improved performance compared to its predecessor, ER 1.6. This enhancement is observed across three control modes: real VLA (Vision-Language-Action), sim VLA (simulated Vision-Language-Action), and human teleoperation. These improvements underscore the model's enhanced ability to interpret visual and linguistic input into motor control, leading to more effective robot actions.
Developer Access and Practical Demonstrations
Developers can access Gemini Robotics ER 2 through the Gemini API and Google AI Studio. Google DeepMind has showcased the model's capabilities by orchestrating a Boston Dynamics Spot robot to fetch objects based on natural language commands. This demonstration highlights the practical application of ER 2 in enabling robots to understand and respond to human instructions in dynamic environments.
Conclusion
The release of Google DeepMind's Gemini Robotics ER 2 on July 30, 2026, marks a notable advancement in embodied AI for robotics. By providing a sophisticated reasoning model that supports multi-step planning, real-time tracking, and multi-robot collaboration, DeepMind is enabling more complex and autonomous robotic systems. Its public availability offers developers new tools to create advanced physical AI agents.
Sources
Was this article helpful?
Found outdated info or have suggestions? Send us a note.
