Google Unveils Gemini Robotics ER 2

Google has released Gemini Robotics ER 2, a high-level 'Embodied Reasoning' model that helps robots understand the physical world and plan multi-step tasks. This model serves as the robot's 'brain,' conversing with humans and understanding the physical environment while passing execution commands to lower-level VLA (Vision-Language-Action) models.

Gemini Robotics ER 2 delivers significantly improved performance over its predecessor, ER 1.6. Key features include:

• Real-time reasoning and execution: Through the Gemini Live API, latency is minimized, enabling flexible orchestration where robots can think about the next step while performing an action.
• Self-correction and adaptation: Through continuous video feed, it tracks its own progress, self-corrects when errors occur, and precisely identifies when to move on to the next step.
• Tool use and collaboration: It supports calling external tools such as Google Search, as well as multi-robot collaboration, where multiple robots work together to complete complex workflows.

This model is now available to developers through the Gemini API, Google AI Studio, and the Gemini Enterprise Agent Platform, and was demonstrated combined with Boston Dynamics' Spot, fetching objects using only natural language commands.