Alphabet’s Google rolls Gemini deeper into robotics with ER 2 that can use search to plan real-world tasks

Google is pushing its Gemini artificial intelligence further out of the screen and into the physical world, announcing Gemini Robotics ER 2, a new AI model positioned as a reasoning engine for robots. In a market report carried by Yahoo Finance, Google described Gemini Robotics ER 2 as a high-level reasoning model that can help robots plan complex tasks, rather than simply follow fixed instructions. The company also said the approach is meant to allow robots to work across different robotic systems, suggesting the reasoning layer is designed to be reusable even as hardware and controls differ. The company’s pitch goes beyond autonomous action. Google said the model can tap Google Search for real-world tasks, implying robots may be able to incorporate information gathered from the web, product databases, maps, or other search-indexed sources when deciding what to do next. That matters because many robotics challenges are not purely mechanical, they are “know-what” and “know-how” problems, where a robot must select an appropriate procedure or identify the right steps for an unfamiliar environment. The announcement comes as large technology firms accelerate their efforts to build what is often called “physical AI,” meaning systems that can operate in the real world, not just generate text, images, or code. In that race, companies are trying to pair large language models and reasoning systems with perception, motion control, and robot-specific tooling, which has historically been difficult to scale across fleets and manufacturers. What Google did not spell out in the published report is likely as important as what it did. The description does not provide performance metrics, such as success rates in particular tasks, latency figures, or benchmarking results against prior robotics models. It also does not say which robot platforms will be able to use ER 2, whether Google is offering it to third parties, or what parts of the robotics stack the reasoning model replaces versus augments. For Alphabet, Gemini Robotics ER 2 fits a broader strategy of expanding Gemini beyond chat and into systems that can interpret goals, plan steps, and integrate information sources. If robots can indeed use search as an input to planning, it could also make the AI more adaptable, because search indexes change and can capture new information without retraining the model for every scenario. At the same time, that dependency raises new questions for robotics deployments, including how reliably the model translates search-backed information into safe, correct actions in dynamic physical environments.

In the near term, investors and industry watchers will likely focus on whether Google follows this announcement with more specifics about partnerships, pilot deployments, and measurable results. The next markers to watch include which robotics systems ER 2 is demonstrated with, what categories of tasks it can handle reliably, and whether Google discloses how it keeps robots within safety and operational constraints when acting on information derived from search.

keyFacts