Intelligent Revolution: Google DeepMind's Dual-Model AI System Empowers Robots with Reasoning and Search Capabilities
Abstract
Google DeepMind recently unveiled its new Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 models, marking AI technology's first true step into the world of physical robotics. These two models feature a dual-architecture design, enabling robots not only to execute commands but also to reason, plan, and proactively use tools, achieving more advanced physical world interaction capabilities.
Breakthrough Dual-Model Architecture
Gemini Robotics-ER 1.5 and Gemini Robotics 1.5 employ an innovative dual-architecture design, working in synergy to form a powerful robotic AI system. Gemini Robotics-ER 1.5 acts as the "high-level brain" responsible for planning and physical space decision-making, while Gemini Robotics 1.5 translates visual and language understanding into concrete actions. This architecture enables robots to perform complex multi-step tasks, such as sorting trash according to local recycling rules found online, or packing luggage based on weather forecasts.
Significant Shift from Reactive to Reasoning
Carolina Parada, Head of Robotics at Google DeepMind, noted that the new system allows machines to "think multiple steps ahead," rather than just focusing on a single step. Unlike previous generations that could only execute simple commands, the new system brings robot behavior closer to human thinking – planning first, then acting.
Cross-Robot Learning Capability
Another revolutionary change brought by this update is the "cross-robot learning" feature. DeepMind demonstrated that skills learned by the dual-arm ALOHA2 robot can be applied to Franka robots or Apptronik's Apollo humanoid robot without additional setup. This means the same set of models can control different types of robots and transfer experiences, significantly accelerating the pace of robotics development.
Intelligent Web Search and Tool Use
Gemini Robotics-ER 1.5 can evaluate complex challenges, natively call tools (such as Google Search) to find information, and create detailed step-by-step plans to overcome problems. This allows robots to find information and apply it to practical tasks much like humans, greatly expanding their range of capabilities.
Enhanced Safety and Transparency
DeepMind placed safety at the core of its development process. Both models are equipped with safeguards and interpretability features, making robot behavior more predictable and explainable. Google also released an upgraded version of the ASIMOV benchmark for evaluating and improving semantic safety, with Gemini Robotics-ER 1.5 demonstrating state-of-the-art performance in safety assessments.
Developer Availability
This update is now available to developers: Gemini Robotics-ER 1.5 can be tested via the Gemini API in Google AI Studio, while Gemini Robotics 1.5 is currently available only to select partners. This will foster broader innovation and application development in robotics.
A Major Milestone Towards General Artificial Intelligence
Google DeepMind describes this release as a significant step towards solving Artificial General Intelligence (AGI) in the physical world. By introducing proactive capabilities, Google is moving beyond models that merely respond to commands, creating systems that can truly reason, plan, proactively use tools, and generalize.
Industry Impact and Future Outlook
This breakthrough technology is poised to reshape multiple industries. In healthcare, assistive robots could provide help tailored to different patient needs; in home environments, they could become smarter personal assistants. While robotics technology is still in its early stages, this release indicates that the future is not just about smarter AI, but about AI that can live and act alongside us.
As Google DeepMind continues to advance the integration of robotics and artificial intelligence, we can expect more innovative applications to emerge, further bridging the gap between virtual intelligence and the physical world.