Boston Dynamics' Atlas has 4 fingers. Tesla's Optimus has 5. Here's why ...

HMHOctober 7, 2026

Boston Dynamics deliberately cut the pinky from Atlas's commercial hand after physical experiments and simulations showed 4 fingers matched the dexterity of 5 for factory tasks. Tesla chose 5 for Optimus to work with human-built tools. The difference isn't a hardware limitation ... it's two different theories of what humanoid robots are for.

Boston Dynamics tested whether Atlas needed a pinky. The answer was no — and the decision to cut it reveals something important about how the two leading humanoid programs think about their robots' futures.

When Boston Dynamics unveiled the new Atlas hands in early October, the team was specific about why: Alberto Rodriguez, who leads Atlas behavior manipulation technology, told reporters the team worked through physical experiments, simulations, and 3D-printed mockups of different finger counts and thumb configurations. Four fingers turned out to match the dexterity of five for factory tasks. The pinky added mechanical complexity without proportional capability.

The result: 13 degrees of freedom, 4 fingers including a thumb with 4 DOF, remaining fingers with 3 DOF each. The hand can carry a loaded mini-fridge. It can insert a drill bit and pull the trigger, tighten a small nut by rolling it between fingertips, and spin a drumstick. It's designed for what Atlas actually has to do, not for how human hands are shaped.

What Hyundai is actually building two of

The detail that clarifies this most is one that didn't get much attention: Hyundai showed two Atlas variants at CES 2026. An Atlas Prototype, used for R&D — 56 degrees of freedom, five fingers. And an Atlas Product Model, the commercial factory unit — four fingers, factory-optimized.

This isn't a case of Hyundai not being able to build a five-finger hand. They built one. They looked at it alongside the four-finger version, evaluated what factory tasks actually require, and chose the simpler design for the commercial product. Hyundai Motor Group Vice Chairman Chang Jae-hoon was direct about it: "Going from four to five fingers isn't a problem, but we considered the most suitable direction and cost for factory use."

That's a deliberate bet on specificity over generality for the commercial product.

Tesla's opposite reasoning

Tesla chose five fingers for Optimus Gen 3, and the reasoning runs the other direction. Optimus is designed to work in environments built by and for humans: using the same tools, handles, switches, wiring harnesses, and equipment that workers use. Five fingers aren't optional if the goal is full human tool compatibility without infrastructure modification.

The Gen 3 hand architecture reflects the ambition: 25 actuators per arm, with all motors moved to the forearm, running cable tendons distally. This is biomimetically close to how human forearm muscles actually control fingers. More degrees of freedom, more tactile sensing, a target of 3,000+ discrete manipulation tasks versus Gen 2's roughly 500. The hand is expensive and complex because the goal is expensive and complex.

Tesla's frame for Optimus is a general-purpose household and factory robot that can do what a human does, anywhere a human would work. Five fingers follow from that premise.

Two theories of the market

The finger count is a proxy for a deeper disagreement about where humanoid robots go to work and what "working" actually means.

Boston Dynamics' position: the near-term commercial opportunity is factory tasks that are specific, repetitive, and well-defined. Design the robot's hardware around those tasks. Cut what doesn't contribute. Train more easily on what remains, because the simulation matches reality better with simpler mechanics.

Rodriguez made the simulation point explicitly: the four-finger design is partly about reinforcement learning. Fewer parts means the sim-to-real gap is smaller. The hand that can be accurately simulated in training is the hand that performs reliably on the floor.

Tesla's position: the largest addressable market for humanoid robots is everything a human does. Building a narrowly optimized factory robot is leaving value on the table. Optimus should work in a car plant, an Amazon warehouse, and eventually a house. Five fingers are the foundation for all three.

Both are coherent.

Agility Robotics, for what it's worth, also went factory-specific: Digit 5 is defined by 10 warehouse use cases, not general-purpose manipulation. Boston Dynamics and Agility are both making the focused bet.

Tesla is making the generalist one.

The 2028 test

Atlas's commercial deployment begins in 2028 at Hyundai and Kia plants in Georgia.

The four-finger hand will have to perform at scale, across multiple tasks, with real economic stakes. If it does, the focused bet will look smart. If the four-finger design runs into tasks it simply cannot handle — ones that a pinky would have solved — the decision will look like a premature optimization.

What's interesting about the Hyundai choice?

It wasn't a limitation. They tested five fingers. They built a five-finger research model. And they shipped a four-finger commercial product anyway.

We'll eventually see which direction makes the most sense.