Google DeepMind has taken a significant leap in AI-powered robotics with the launch of Gemini Robotics 2, a vision-language-action (VLA) model that enables complete control of humanoid robots — from feet to fingertips. Announced on July 30, 2026, the model represents the most ambitious evolution of the Gemini Robotics series, moving beyond table-top manipulation tasks to movements requiring full-body coordination.
The key innovation in Gemini Robotics 2 is its unified VLA architecture. Unlike previous approaches that segmented robotic control into separate modules — vision, planning, motor execution — the new model integrates everything into a single system that receives visual and linguistic input and produces motor commands directly. This allows a robot to not only "see" and "understand" an instruction like "pick up the blue box on the top shelf," but also plan and execute the complete sequence of movements — bending, reaching, rotating the torso, grasping — without human intervention.
The technical leap is substantial. While the original Gemini Robotics focused on arm control and object manipulation on tabletop environments (already impressive), Gemini Robotics 2 expands to full humanoids. The model manages dynamic center of mass during movements like crouching to pick something off the floor or balancing while carrying heavy objects. It also supports bi-arm robots, opening possibilities for tasks requiring two-arm coordination, such as opening a door while carrying a box.
DeepMind also announced advances in fine dexterity with Gemini Robotics ER (Enhanced Research), which adds five-finger manipulation capability — something The Next Web noted still lags behind human dexterity but represents remarkable progress from simple grippers. There's also Lyria 3.5, integrating musical capabilities, but the primary focus remains physical robotics.
The implications for industry are enormous. Logistics companies like Amazon and FedEx, which already use robots in warehouses, could benefit from humanoids that navigate tight spaces and perform varied tasks without reconfiguration. In manufacturing, full body control opens the door for robots operating on assembly lines traditionally designed for humans — without requiring factory floor modifications.
The announcement also positions Google DeepMind on a collision course with companies like Tesla (developing Optimus), Figure AI (with Figure 02), and Unitree (with H1). The difference is that Google isn't building the hardware — it's providing the "brain." This could prove more scalable: instead of competing in mechanical design, Google bets its AI model will be adopted by multiple robot manufacturers, functioning as the "Android of robots."
However, the path to widespread adoption still faces significant hurdles. The cost of humanoid robots remains prohibitive for most commercial applications, and safety concerns — both physical (an 80 kg out-of-control robot) and digital (hacking of robotic systems) — need addressing. Google's massive AI investment base, estimated at over $50 billion annually, suggests the company is prepared for the long haul. The question remains whether the market is ready to embrace "full-body intelligence" robots — or if this is still a technology in search of a problem. The announcement also puts pressure on competitors like Figure AI and Tesla Optimus, which have been developing their own humanoid robots with varying degrees of autonomy. Figure AI recently demonstrated its Figure 02 performing warehouse tasks, while Tesla's Optimus remains in early prototype stages. Google DeepMind's approach — providing the AI brain rather than building the body — could accelerate the entire field by allowing multiple hardware manufacturers to leverage the same advanced control software. The company has indicated it will make Gemini Robotics 2 available to select research partners, suggesting a controlled rollout rather than an open release.
Sources: Google DeepMind, Wired, The Next Web
✓ Independent sources cross-checked and verified before publishing
