NeuralStackGoogle AI is drawing a clear line from its early breakthroughs in games and biology to a new generation of systems that act in the physical world. At the top of its latest slate of projects is Gemini Robotics 2, which the company describes as bringing whole body intelligence to robots. While the technical details remain under wraps in this snapshot, the positioning makes it clear that Google DeepMind now sees embodied reasoning and control as a core frontier for its Gemini models.
The robotics work sits alongside a growing portfolio of AI aimed squarely at scientific discovery. In September 2025, Google Research highlighted Empirical Research Assistance (ERA), a system designed to help move work from Nature‑level publications toward catalyzing new computational discoveries. ERA is framed less as a single tool than as a research assistant that sits in the loop with scientists, suggesting Google wants its AI to be part of how hypotheses are tested and refined, not just how papers are written.
That same ambition runs through Gemini for Science, announced in May 2026 and backed by Google Labs, Google DeepMind and Google Research. Rather than one monolithic product, Gemini for Science is presented as a collection of AI experiments and tools for a new era of discovery. By bundling these efforts under the Gemini banner, Google is trying to turn its general‑purpose models into a flexible platform that can adapt to different scientific domains, from physics and biology to materials and beyond, even if the exact workflows remain deliberately broad at this stage.
Google DeepMind, Google Cloud, Google Labs and Google Research are also collaborating on Co‑Scientist, introduced in May 2026 as a multi‑agent AI partner to accelerate research. The concept is to move beyond a single chatbot into a coordinated team of AI agents that can help with everything from literature review to experiment planning. That multi‑agent framing matters: it signals that Google is betting on structured collaboration between specialized AI systems as the way to scale up complex projects, rather than relying on one giant model to do everything.
Robotics remains a recurring theme across Google’s timeline of breakthroughs. Before Gemini Robotics 2, Google DeepMind rolled out Gemini Robotics ER 1.6 in April 2026, describing it as a way to power real‑world robotics tasks through enhanced embodied reasoning. Taken together, ER 1.6 and Gemini Robotics 2 suggest an iterative path: first improving how robots reason about their environment, then tackling the broader challenge of whole‑body coordination and control. Google’s emphasis on real‑world tasks hints at an intent to move beyond lab demos toward systems that can operate in more varied and practical settings.
Parallel efforts show how Google is using its AI to deepen our understanding of everything from games to galaxies. A March 2026 DeepMind retrospective on "From games to biology and beyond: 10 years of AlphaGo’s impact" connects the company’s early triumph in Go to downstream advances like AlphaFold and newer biology work. In October 2025, Google Research described teaching Gemini to spot exploding stars with just a few examples, while another project focused on using AI to perceive the universe in greater depth. These astrophysics‑oriented systems sit alongside tools like WeatherNext 2, Google’s most advanced weather‑forecasting model as of November 2025, underscoring how the same core AI approaches are being stretched across multiple scientific disciplines.
Medical and environmental applications feature prominently as well. Google Research has detailed how AI can improve breast cancer detection in the UK, and separate work on DeepSomatic focuses on using AI to identify genetic variants in tumors. On the ecological side, Google Research has outlined how it is helping preserve the genetic information of endangered species with AI, and how the FireSat satellite constellation is beginning to deliver first images of wildfires. Together with tools like AlphaProteo and the continuing evolution of AlphaFold, these projects show Google positioning its AI stack as an end‑to‑end pipeline: from detecting disease and cataloging biodiversity to generating novel proteins for health research.
Why this matters
While many AI headlines focus on chatbots and coding assistants, Google’s latest research slate points to something broader: a push to turn foundation models into engines for real‑world discovery and action. Robotics projects like Gemini Robotics 2 and ER 1.6 hint at a future where AI systems don’t just reason, but also move and manipulate the world with whole‑body intelligence. Meanwhile, multi‑agent tools such as Co‑Scientist and platforms like Gemini for Science and ERA show Google trying to weave AI directly into the fabric of how science is done, from initial ideas through to experimental validation. If these efforts mature, they could reshape not just consumer products but the workflows of labs, observatories, hospitals and conservation projects globally.
The research timeline also reveals how Google is building supporting infrastructure around its flagship models. Projects like Genie 3, described as a general‑purpose world model that generates a diversity of interactive environments, give Gemini‑based agents simulated spaces in which to play, reason and learn before they graduate to the physical world. Hardware‑adjacent efforts such as the Coral NPU, billed as a full‑stack platform for edge AI, indicate a parallel push to get these models running closer to sensors and devices. And in more human‑centric domains, tools like SensorLM, which "learns the language" of wearable sensors, and Aeneas, which transforms how historians connect the past, suggest Google is testing how far its AI can go in interpreting both biological and cultural signals.
Google’s catalog of breakthroughs over 2024–2026 reads less like a set of disconnected demos and more like an emerging ecosystem. AlphaFold and AlphaProteo anchor the biology side, WeatherNext 2 and astrophysics‑focused Gemini models expand into climate and cosmology, FireSat and endangered‑species genomics tackle environmental risk, while robotics and multi‑agent research aim to close the loop between digital reasoning and physical action. As Gemini Robotics 2 takes center stage, the next phase will be proving that "whole body intelligence" in robots and multi‑agent Co‑Scientists can deliver sustained, measurable gains in scientific output and practical capability, not just impressive videos. How Google DeepMind, Google Research, Google Labs and Google Cloud execute on that promise will determine whether these projects remain breakthroughs on paper or become the backbone of everyday tools in labs, factories and the field.