Google's Gemini Robotics-ER 1.5 Achieves State-of-the-Art in Embodied Reasoning, Now Available to Developers

Google has announced the launch of Gemini Robotics-ER 1.5, its first widely available robotics model, achieving state-of-the-art (SOTA) performance in Embodied Reasoning tasks. Logan Kilpatrick stated in the tweet, "> "Introducing our first widely available Robotics model: Gemini Robotics-ER 1.5 🤖 It is currently SOTA on a set of Embodied Reasoning tasks and can be used used directly through the Gemini API or Google AI Studio!"" This release on September 25, 2025, makes advanced AI agent capabilities accessible to developers globally.

Serving as a robot's "high-level brain," Gemini Robotics-ER 1.5 is a Vision-Language Model (VLM) adept at spatial understanding, complex task planning, and progress estimation. It can natively integrate digital tools, including Google Search, to orchestrate multi-step tasks by breaking them into logical sub-components. The model's verified SOTA performance across various spatial understanding benchmarks highlights its sophisticated ability to interpret and reason about the physical environment.

The model's availability to developers via the Gemini API and Google AI Studio is poised to accelerate the creation of more capable and versatile robotic applications. Gemini Robotics-ER 1.5 is particularly suited for daily tasks that demand contextual understanding and sequential execution, such as sorting items according to specific guidelines. Its flexible "thinking budget" allows developers to fine-tune the balance between low-latency responses for simple actions and enhanced accuracy for intricate reasoning challenges.

This introduction is a key component of Google's broader strategy to foster intelligent, general-purpose robotics, working in conjunction with the Gemini Robotics 1.5 model, a Vision-Language-Action (VLA) system currently offered to select partners. While Gemini Robotics 1.5 translates visual data into direct motor commands, ER 1.5 focuses on the cognitive aspects of reasoning and planning. This dual approach signifies a crucial stride toward realizing Artificial General Intelligence (AGI) within physical domains.

Google has underscored its commitment to responsible AI development, integrating enhanced semantic safety and physical constraint awareness into Gemini Robotics-ER 1.5. The model is engineered to identify and reject operational plans that might violate predefined physical limits or safety protocols. Despite these built-in safeguards, the company stresses the critical responsibility of developers to implement comprehensive safety engineering and adhere to best practices for real-world robotic deployments.