.svg){kind=spacer}
Details
- Google for Developers has introduced Gemini Robotics-ER 1.5, a specialized addition to the Gemini model family designed for real-world decision-making by robots.
- The Preview release is now accessible via Google AI Studio for select researchers and robotics teams, offering API integration.
- New capabilities feature advanced agentic behaviors for planning complex tasks, rapid multimodal perception, improved safety filters, and a configurable "thinking budget" to manage reasoning depth versus computing costs.
- The model is trained on extensive robot-manipulation datasets and physics simulations, enabling generation of executable motion plans, not just textual commands.
- Gemini Robotics-ER 1.5 runs on Google's Cloud TPU v6e with the ability to export streamlined policies to EdgeTPU or ARM chips, supporting various mobile, warehouse, and assistive robot platforms.
- A safety guardrail layer enforces joint-limit, speed, and collision constraints, addressing known risks in embodied-AI safety.
- Python and ROS 2 SDKs are included, with updates on general release, pricing, and on-device compatibility forthcoming.
Impact
This launch extends Google's lead in embodied AI over competitors like NVIDIA and OpenAI, raising the stakes in the robotics software market. The model’s safety and compliance features position Google strongly amid evolving regulatory requirements. Its hybrid cloud-to-edge support and flexible development tools could accelerate adoption, while the outcome of this preview will shape future investments in Gemini’s AI roadmap.
