Google Releases Gemini Robotics 1.5: Advanced AI Models for Physical World Automation
- Nikita Silaech
- Sep 30
- 1 min read

Google has released Gemini Robotics 1.5, a breakthrough AI system that enables robots to perform complex, multi-step tasks in physical environments through advanced reasoning and planning capabilities.
Technical Architecture:
Dual-Model Framework: Combines Gemini Robotics-ER 1.5 (embodied reasoning model) for high-level planning with Gemini Robotics 1.5 (vision-language-action model) for direct motor control and execution.
Agentic Thinking: First vision-language-action model that thinks before acting, generating internal reasoning sequences in natural language to break down complex tasks into executable steps.
Cross-Embodiment Learning: Demonstrates remarkable ability to transfer learned behaviors between different robot types without model specialization, accelerating skill acquisition across platforms.
Tool Integration: Native capability to call external tools like Google Search and third-party functions, enabling robots to access real-time information for context-dependent tasks.
Performance Benchmarks:Â Gemini Robotics-ER 1.5 achieves state-of-the-art results across 15 academic embodied reasoning benchmarks, including Point-Bench, ERQA, and RoboSpatial-VQA. The system successfully demonstrates complex scenarios like location-based waste sorting that requires internet research, object recognition, and multi-step execution.
Safety Framework:Â Implementation includes comprehensive safety measures through the upgraded ASIMOV benchmark, featuring semantic reasoning for safety assessment, alignment with Gemini Safety Policies, and integration with low-level collision avoidance systems.
Research Significance:Â The release represents a foundational advance toward artificial general intelligence in physical environments, moving beyond reactive command execution to systems capable of autonomous reasoning, planning, and tool usage.
Availability:Â Gemini Robotics-ER 1.5 is available to developers via the Gemini API in Google AI Studio, while Gemini Robotics 1.5 remains available to select partners for specialized applications.
Source: Google Deepmind Blog