Google's Humanoid Robot Claims: 80% Real, 20% Unverified

Google DeepMind unveils Gemini Robotics 2 — a major step toward general-purpose humanoid robots.
Gemini Robotics 2 enables full-body control, from walking, crouching, and balancing to precise five-finger object manipulation.
The system is powered by three models:
• Gemini Robotics 2 — Converts vision and natural language instructions into real-world robot actions.
• Gemini Robotics ER 2 — Handles long-horizon planning, tracking hundreds of decisions across multi-minute tasks and coordinating multiple robots.
• On-Device 2 — Runs entirely on the robot and adapts to new hardware with fewer than 200 demonstrations collected in just a few hours.
A single Gemini Robotics 2 checkpoint successfully controlled both Apptronik Apollo humanoids and the Franka Duo robot, demonstrating strong cross-platform generalization.
Current performance still leaves room for improvement:
🔹 Whole-body task success: 45.7%–76.3%
🔹 Five-finger manipulation: 32% (dustpan use) to 92% (unscrewing a light bulb)
The significance isn’t perfect accuracy—it’s a foundation model that can transfer across different robot bodies while combining perception, reasoning, planning, and control in one system.

Fact-check: the post is largely accurate, confirmed by multiple sources (Google DeepMind's own blog, Bloomberg, SiliconANGLE, Robotics and Automation News). Here's the breakdown:

Confirmed accurate:

• Google DeepMind did unveil Gemini Robotics 2 (announced July 30, 2026), enabling whole-body humanoid control — walking, crouching, balancing, and five-finger manipulation, expanding beyond the prior model's upper-body-only control.While previous models controlled the humanoid's upper-body to achieve table-top tasks, Gemini Robotics 2 expands physical AI into whole-body motions, controlling entire humanoid robots for the first time
• The three-model structure is correct: Gemini Robotics 2 is the most advanced vision-language-action model that converts vision and language input into motor control, paired with an embodied reasoning model (ER 2) and an on-device model.
• ER 2 handles long-horizon, multi-step planning and multi-robot coordination — it "enables multiple autonomous machines to collaborate on a task" and "can automate chores that comprise hundreds of steps."
• On-Device 2's few-hours/adaptation claim checks out: Google DeepMind said it can adapt to entirely new robot bodies with just a few hours of data, typically fewer than 200 examples.
• Apptronik's Apollo 2 (not "Apollo," as the LinkedIn post says) is the primary demo humanoid: the system is being demonstrated on Apptronik's Apollo 2 humanoid robot.

Needs correction:

• The post says "Apptronik Apollo" — the actual robot is Apollo 2, a distinction several outlets specifically note as new hardware tied to this release.
• I could not verify the "Franka Duo" claim or the specific percentage figures (45.7%–76.3% whole-body success; 32% dustpan/92% lightbulb) in the sources I checked. These numbers are plausible (DeepMind's blog does discuss task success rates and dexterity benchmarks), but I'd treat them as unverified until confirmed against DeepMind's technical report or blog post directly — happy to dig into the primary source if you want those numbers checked precisely.

Overall: The core narrative — full-body humanoid control, three-model architecture, cross-platform generalization, imperfect but promising success rates — is real and well-sourced. The main flag is "Apollo" vs. "Apollo 2," and the specific numeric benchmarks need direct verification against DeepMind's technical documentation.