The hand is the hard part, and so is telling it what to do

Twenty-two degrees of freedom in a hand, and a factory ordering hundreds of units. The bottleneck is no longer the hardware — it is specifying the task without an engineer in the building.

Gemini Robotics 2 drives the five-fingered, 22 degree-of-freedom SharpaWave hand on Apptronik's Apollo 2, with DeepMind claiming full body control.

Why the hand is different

Arm positioning is essentially solved — well-understood inverse kinematics, decades of practice. A 22 degree-of-freedom hand doing manipulation is high-dimensional, contact-rich control where the model has to reason about force and slip rather than trajectory. It is a different problem wearing the same word.

Full body control also reaches further down the stack than the recent consensus. Embodied foundation models have mostly been positioned as planners above dedicated controllers, precisely because low-level control needs deterministic latency and hard safety guarantees. Claiming the whole body reopens that question.

Running on a third party's humanoid with a third party's hand is what makes it credible. Cross-vendor demonstrations are much harder to stage than in-house ones.

The other end of the same bottleneck

Tacta unveiled a three-part system built around users teaching dexterous skills rather than engineers programming them, while Schaeffler committed to hundreds of Humanoid units across its factories.

Teaching versus programming is an economic argument before it is a technical one. A robot needing an engineer per task has a cost structure that only works at enormous volume. One a line worker can demonstrate a task to has a completely different deployment curve.

And hundreds of units is an order, not a pilot. The sector's own analysis has been consistent that pilot-to-fleet is where nearly every humanoid programme has previously died.

The convergence

Both stories describe the same constraint from opposite ends. Hardware is now adequate for a growing set of real tasks. What remains scarce is a way to specify the task that does not require the vendor's engineers on site.

Whoever solves specification, rather than dexterity, takes this market.

The Robot Report — Google DeepMind says Gemini Robotics 2 enables full body control → · The Robot Report — Schaeffler to deploy hundreds of Humanoid robots in its factories → · The Robot Report — Top 10 robotics stories of July 2026 →