Gemini Robotics 2 Gives Humanoid Robots Full-Body Control

A New Kind of Robot Brain

Google DeepMind has introduced Gemini Robotics 2, an AI model designed to control humanoid robots holistically — from balance and gait through to the fine motor skills needed to manipulate small objects with their fingertips 12. Rather than treating locomotion, grasping, and coordination as separate problems handled by different systems, the new model unifies them into a single foundation model that governs a robot's entire body at once 14.

Reporting on the release describes robots demonstrating everyday dexterity tasks that have long stumped machines, such as tying up a trash bag or unscrewing a lightbulb, alongside more dynamic capabilities like maintaining balance while walking and performing delicate handling tasks 74. The model is also said to support multi-robot teamwork, letting separate machines coordinate on shared tasks, and it reportedly incorporates safety-aware behavior so robots can operate more predictably around people 24. Coverage frames this as part of a broader push toward "robot foundation models" — general-purpose AI systems that can be deployed across many robot bodies and tasks rather than being hand-tuned for a single narrow function, echoing wider industry momentum around foundation models for industrial robots, cobots, drones, and service robots 9.

Why Full-Body Control Matters

Humanoid robotics has historically split into silos: separate teams and models for walking, balancing, and manipulating objects. A single model spanning "feet to fingertips" suggests progress toward robots that can generalize across tasks without needing bespoke programming for each new skill 12. That matters because dexterity — precise, adaptive use of hands and fingers — has been one of the hardest unsolved problems in robotics, arguably harder than locomotion itself.

A Sector Facing Turbulence Elsewhere

Even as the underlying AI advances, the humanoid robot business faces real-world friction. Elon Musk has tempered expectations for Tesla's robotaxi and Optimus humanoid ambitions, citing what he called a "substantial" engineering challenge still ahead, a reminder that impressive demos don't guarantee near-term commercial deployment 3.

Meanwhile, humanoid robots are running into regulatory and logistical hurdles far removed from AI labs. United Airlines has joined Southwest and Delta in barring humanoid and animal-like robots from passenger cabins and checked baggage, reflecting airline caution about unfamiliar, potentially battery-heavy or moving equipment on flights 5.

At the geopolitical level, the Trump administration has banned imports of Chinese-made humanoid robots, citing national security risks 68. The move is notable given China's dominance of humanoid robot manufacturing, with analysts estimating its global market share at around 85% 6. That dependency underscores the stakes for U.S. firms like Google and Tesla as they race to build both the software and hardware for humanoid robots domestically, even as trade barriers reshape who can supply the physical machines these new AI models are meant to control.