Atlas Gets a 13-Joint Hand and a 100,000-a-Year Production Goal; Price and Autonomy Stay Blank

The number to dispute in Boston Dynamics' new Atlas hand is not the 13 joints but the 100,000. Alberto Rodriguez, the company's director of robot behavior, said in IEEE Spectrum that "what we're working on right now is figuring out what needs to change in the fine details of our design if we want to make 100,000 of these hands a year." It is a goal, and Rodriguez says the design work to reach it is still going on.

IEEE Spectrum's Evan Ackerman published the hand's details on October 1, 2026, and Robotics 24/7 followed on October 2. Neither gives a price, and neither says whether the demonstrations were autonomous or teleoperated.

What changed from 7 joints to 13

The previous Atlas hand had 7 degrees of freedom. The new one has 13: a thumb with 4 and three fingers with 3 each, according to Robotics 24/7. Spectrum describes three fingers and a thumb, and says the pinky was removed after the team "spent a day with their pinkies taped to their ring fingers" to see what they lost.

The fingers can splay, which lets one finger work against another and lets the hand hold a tool's trigger handle. Spectrum says reinforcement learning could exploit motions beyond the human range.

1. Previous Atlas hand7 degrees of freedom
2. New Atlas hand13 degrees of freedom

Source: IEEE Spectrum, October 1, 2026; Robotics 24/7, October 2, 2026

Direct drive and a replaceable pack

The actuators sit inside the joints in a direct-drive arrangement. The transmission can be back-driven, so the motors give way to force and contact instead of fighting it. Each actuator pack is one replaceable unit, and no tendons or cables cross a joint, Spectrum reports.

That is the part of the design that bears on the 100,000 figure. Rodriguez said "hands are a ruthless design trade-off," and that "many of the current designs are giving up on things like reliability and manufacturability." A hand with no cable runs and one replaceable actuator pack is meant to be cheaper to build and quicker to repair, which is the argument Boston Dynamics is making against more humanlike hands.

The argument is the company's. Spectrum calls the hand "awfully clunky" in appearance and runs no benchmark, no outside evaluation and no critic. The headline's claim that it "may outperform" humanlike designs rests on the company's goals and the author's framing.

The arithmetic behind 100,000 hands

A humanoid has two hands. At 100,000 hands a year, that is hardware for 50,000 robots a year, before spares and before counting any robot that is not an Atlas. That is my arithmetic, not a Boston Dynamics figure, and the company has not given a production date or a robot volume.

The same goal also implies 1.3 million degrees of freedom a year (13 per hand, 100,000 hands), each of which needs an actuator that survives a factory shift. Rodriguez framed the open problem himself: the team has not yet settled which fine details of the design must change to reach that volume, so the 100,000 describes an intent rather than a tooling plan.

For scale, this site reported on Industrial Next's claim that factory robots can switch tasks in minutes; hand volume of that order assumes a buyer base well beyond a few evaluation sites.

Item	What the sources say	Source
Joints per hand	13, up from 7	IEEE Spectrum
Digits	Three fingers and a thumb	IEEE Spectrum
Load shown	Loaded minifridge over 100 lb	Robotics 24/7
Production goal	100,000 hands a year, design in progress	IEEE Spectrum
Price	Not given	Both
Demo autonomy	Not given	Both

Demos are not deployments

Robotics 24/7 says the hand carried a loaded minifridge weighing over 100 lb, and that the new generation of Atlas has a payload above 100 lb. Spectrum lists tool use including drills and welding torches among the intended tasks. The pairing of a tool demo with a control mode no source gives is the same gap that leaves any humanoid demo hard to read; an earlier piece on a driver-assistance claim that stayed unconfirmed shows how long such gaps can last.

The display model above is an Atlas shown at a Hyundai exhibition in July 2026; the new hand is not visible in the frame.

What would change my read

Three disclosures would settle it. A unit cost at a named volume would test the manufacturability claim. A statement of which demonstrations ran without a teleoperator would test the training claim. And a hand-hours or cycles-to-failure figure would test the reliability claim that Rodriguez puts at the centre of the design. Until then the supportable summary is a hand with twice the joints, no pinky and a volume target that the company says it has not yet engineered for. Robotics 24/7 adds that the hand is meant to be mass-manufactured and repaired at low cost, but gives no figure for either, so the cost claim is as open as the volume one.