Google DeepMind Robots Now Learn from the Web, Handle Complex Real-World Tasks
In a groundbreaking advancement for artificial intelligence and robotics, Google DeepMind has unveiled its latest AI-powered robotic models that not only perform complex physical tasks but can also use the internet independently to gather information and make informed decisions. This fresh wave of innovation signals a new era where robots are evolving from simple task executors into intelligent agents capable of planning, reasoning, and problem-solving in real-world environments.
The Gemini Robotics Breakthrough
Google DeepMind has introduced two cutting-edge models named Gemini Robotics 1.5 and Gemini Robotics-ER 1.5. These models work in tandem to empower robots with remarkable abilities beyond traditional automation. Gemini Robotics-ER 1.5 functions as the brain, equipped with embodied reasoning abilities—allowing robots to understand their surroundings, think several steps ahead, and use digital tools like Google Search to access real-time information. Gemini Robotics 1.5 acts as the hands and eyes, turning these instructions into physical actions using sophisticated vision and language understanding.
This dual-model approach enables robots to tackle multi-step tasks that require adaptability and proactive problem-solving. For instance, a robot can now sort laundry by colour, pack a suitcase considering varying weather conditions (such as London’s current forecast), or separate waste according to local recycling guidelines fetched dynamically from the web.
From Reactive to Proactive Robotics
Previously, robots were confined to executing rigid, pre-programmed instructions—good at precisely repeating simple tasks but unable to adapt or think beyond the immediate command. As Carolina Parada, Head of Robotics at DeepMind, explains, the update transitions robots from executing “one instruction at a time” to demonstrating genuine understanding and autonomous problem-solving. Robots can now anticipate outcomes, adjust actions in real time, and use external information sources to enhance task effectiveness.
For example, when asked to sort trash, a robot using Gemini Robotics-ER 1.5 searches the internet for location-specific waste disposal policies. It then interprets this data to make the correct decisions on compost, recycling, and trash sorting, ultimately improving efficiency and accuracy in everyday tasks.
Cross-Robot Learning Abilities
One of the most remarkable features of these AI models is their ability to generalise skills across different robotic platforms. Google researchers demonstrated that tasks learned by one robot with specific mechanical arms translated effortlessly to other robots, including humanoids with entirely different configurations. This interoperability means that a skill developed on one robot can be transferred, allowing rapid deployment and scaling of robotic capabilities without redesigning the AI from scratch for every new robotic system.
Such cross-robot skill transfer paves the way for unified AI systems that can control various types of robots, making robotics more flexible, scalable, and cost-effective in industries ranging from manufacturing and logistics to personal assistance and healthcare.
Paving the Way Toward Human-Level AI in Robotics
Google DeepMind views this milestone as a critical step toward Artificial General Intelligence (AGI) capable of human-level reasoning and dexterity in the physical world. By endowing robots with “agentic capabilities”—the ability to plan, reason, and actively use tools—DeepMind is pushing the boundaries of robotics from programmed automata to autonomous intelligent agents.
The rollout includes making the embodied reasoning model (Gemini Robotics-ER 1.5) accessible to developers via Google AI Studio, signaling a potential wave of innovation and new applications powered by these intelligent robotic capabilities in the near future.
Real-World Implications and Future Prospects
This breakthrough introduces vast opportunities for sectors requiring sophisticated robotic assistance. From automating complex manufacturing lines to aiding in elder care and performing household chores, robots embedded with such intelligence could augment productivity and quality of life globally. Smart, web-enabled robots could also pioneer new research methods in labs by autonomously handling complicated experiments or maintenance tasks.
As Google DeepMind continues refining these models and expanding their availability, the day when robots become common, multi-talented helpers that move and think like humans seems closer than ever.
