Gemini Robotics 2 Gives Humanoid Robots Whole-Body Intelligence — Why Auto Factories Matter

Artificial intelligence is starting to move beyond screens and into the physical world. On July 30, 2026, Google DeepMind introduced the Gemini Robotics 2 family, extending robot intelligence beyond tabletop manipulation to whole-body control of humanoid machines.

In Google DeepMind’s demonstrations, Gemini Robotics 2 controls Apptronik’s Apollo 2 as it walks, crouches, bends, reaches for objects and carries them to a destination. For the automotive industry, the bigger question is not simply whether a humanoid can walk. It is whether physical AI can eventually work inside factories that were designed around human movement.

Gemini Robotics 2 controlling an Apptronik Apollo 2 humanoid robot with whole-body motion
Google DeepMind demonstrates Gemini Robotics 2 whole-body control using Apptronik Apollo 2. (Image: Google DeepMind official video)

From understanding instructions to controlling the whole body

Gemini Robotics 2 is a family of three models with different roles. Gemini Robotics 2 is the vision-language-action model that turns visual and language inputs into motor control. Gemini Robotics ER 2 handles embodied reasoning, multi-step planning, progress tracking and multi-robot collaboration. Gemini Robotics On-Device 2 is designed to run locally on robot hardware, reducing reliance on a continuous network connection.

Together, the system separates lower-level physical control from higher-level reasoning and local execution. In demonstrations, Apollo 2 can receive a natural-language instruction, walk toward an object, pick it up and move it to another location. ER 2 can also handle tasks that last several minutes and involve hundreds of decisions, with the ability to re-plan when a step fails.

The dexterity gap is still real

The demonstrations are impressive, but Google’s own evaluation results are an important counterweight to the hype. Apollo 2’s average success rates for picking up objects from a table, floor and shelf ranged from 45.7% to 76.3%. Some five-finger manipulation tasks, such as tying a trash bag, using a dustpan or sealing a ziplock bag, were still successful only around three to four times out of ten.

Google DeepMind also notes that multi-finger dexterity remains challenging and that movement speed still needs improvement. The important advance, then, is that the range of tasks is expanding — not that humanoid robots have already reached the flexibility of skilled human workers.

Why car factories are a natural test bed

Automotive plants have used industrial automation for decades, but conventional robot arms are strongest when position, path and repetition stay predictable. When parts, layouts or production sequences change, manufacturers may need new programming, fixtures or even redesigned workstations.

Humanoids offer a different proposition. Factory aisles, shelves, workbenches and tools are already designed around the human body. If a robot can navigate the same spaces and use similar interfaces, it could potentially handle material movement, sorting, kitting, inspection and other variable tasks without requiring every area to be rebuilt around dedicated automation.

Apptronik already has a real automotive manufacturing connection. Mercedes-Benz entered a commercial agreement with Apptronik in 2024 to explore Apollo in manufacturing, while Apptronik said in 2026 that Apollo 2 data-collection workflows had expanded to partner and customer sites including Mercedes-Benz. That does not mean Mercedes-Benz has deployed Gemini Robotics 2 itself, and the two developments should not be conflated.

Reliability, safety and economics matter more than demos

A factory needs more than an impressive video. Robots have to work for long periods, cope with variation, recover safely from uncertainty and stop appropriately when people enter their workspace. Google says Gemini Robotics 2 adds stronger safety reasoning, including asking for human intervention when uncertain and triggering safe-stop behaviour around people.

The commercial test will ultimately be reliability, speed, maintenance cost, safety certification and return on investment. In the near term, humanoids are more likely to support workers on repetitive, tiring, hazardous or ergonomically difficult tasks than to replace an entire production workforce.

Move Auto's Take

Automotive manufacturing has been automated for decades, but the promise of humanoids is not to turn a conventional robot arm into a human shape. It is to tackle jobs that are too variable to justify redesigning an entire production line around them. Gemini Robotics 2 moves that idea forward, but the real race is about making robots dependable, safe and affordable enough to work every day.

Comments

Popular Stories

F1 Sepang 2026 MyKad tickets reportedly from RM200 — in 2017, entry started at RM58.68

Bugatti Destrier turns the Bolide into a one-off coachbuilt W16 sculpture

Where Did Malaysia’s Tuning Culture Go? From Civic Kids and Silvias to Art of Speed 2026