Atlas Robot Hand: Why It Beats the Human Design

The new Atlas robot hand has no pinky. It has three fingers and a thumb, 13 degrees of freedom, and its motors sit inside the knuckles rather than pulling on cables from a forearm. IEEE Spectrum reported on October 1, 2026 that this design may outperform humanlike hands, and the reason is not that it moves better. It is that it can be built and repaired.

Alberto Rodriguez, Boston Dynamics’ director of robot behavior, put it without decoration: “Hands are a ruthless design trade-off. There’s no way around it, you’re always giving up on something.”

Key Takeaways

• Three fingers plus a thumb, 13 degrees of freedom, roughly double the previous generation’s 7. The pinky was removed deliberately.
• Actuators are embedded directly in the joints in a direct-drive configuration, using back-drivable motors that react to force and contact. There are no tendons or cables to stretch and break.
• Each actuator pack is a single swappable unit, which turns a repair into a parts swap rather than a rebuild.
• The hand can splay its fingers independently against each other, which the article notes exceeds what a human hand can do.
• The real design target is volume. Rodriguez described the team “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.”

What Is Inside the New Atlas Robot Hand

Mechanism first, because the whole story is in where the motors live.

In a human hand, most of the muscles that move your fingers are not in your fingers. They are in your forearm, and they pull on tendons that run through your wrist. Feel your own forearm while you curl a finger and you will find the muscle doing the work several centimetres away from the joint it moves.

That architecture exists for a good reason. Keeping the motors out of the fingers makes the fingers slim and light, which means low inertia, which means they can move fast and fit into tight spaces. It is an elegant solution and evolution arrived at it.

It also has a cost, and the cost is tendons. Tendons stretch. In a machine, cables fray, lose tension, need re-tensioning and eventually snap. The Shadow Dexterous Hand, a long-standing research platform, shows what the human-mimicking approach looks like when engineered: 24 joints with 20 degrees of freedom, driven by 20 DC motors in the forearm, or in the pneumatic version by 20 antagonistic pairs of air muscles, with Hall effect sensors in every joint. It reproduces the human hand’s degrees of freedom and it is a research instrument, not a mass-produced part.

Boston Dynamics went the other way. Put the actuator in the joint. Direct drive, no transmission, no cable. The motors are back-drivable, meaning external force can turn them, which is how the hand senses contact: if something pushes a finger, the motor feels it. Fewer, larger, more powerful actuators replaced many small ones, and each one is a module you can pull out.

The consequences, in order:

Fingers get bulkier. Motors take space, so you cannot have five slim digits. Hence three fingers and a thumb.

Reliability goes up. Nothing stretches. The failure modes that plague tendon-driven hands simply are not present.

Repair becomes trivial. A swappable actuator pack means a technician replaces a unit rather than restringing a hand.

Sensing comes for free. Back-drivable direct-drive motors are their own force sensors, which removes a whole category of fragile fingertip sensors.

And one unexpected gain. The fingers can splay independently against each other, going beyond human range. The design gave up a pinky and picked up a motion your hand cannot make.

Four Hands Compared

	Human hand	Shadow Dexterous Hand	Previous Atlas hand	New Atlas hand
Digits	Five	Five	Fewer than human	Three fingers and a thumb
Degrees of freedom	The reference point	20 actuated, 24 joints	7	13
Where the actuators are	Forearm, via tendons	Forearm, via tendons or air muscles	Not built for volume	Inside the joints
Force sensing	Nerves and skin	Dedicated force or pressure sensors per axis	Varies	The motors themselves, back-drivable
Main failure mode	Injury, fatigue	Cables and tendons	Not designed for mass production	To be established in service
Repairability	Biology	Specialist rebuild	Specialist rebuild	Swap an actuator pack
Built at scale?	n/a	No, research platform	No	The explicit goal, 100,000 a year

The bottom two rows are where the argument lives. Boston Dynamics did not build a better hand in the biological sense. They built a hand that can exist in quantity, which is a completely different objective, and the one that determines whether robots show up in actual workplaces.

What “Beats the Human Design” Actually Means

Careful here, because the headline claim is easy to overread.

The human hand is extraordinary at dexterity, at tactile sensing, at self-repair and at operating for eighty years on food. No robot hand approaches it on those axes. Reinforcement learning work on in-hand manipulation, including OpenAI’s 2018 study using a Shadow Dexterous Hand to reorient objects by vision, demonstrated impressive behaviour and also demonstrated how much machinery and training it takes to get there.

What the new Atlas hand beats the human design on is a different list: manufacturability, serviceability, consistency and one specific range of motion. Those are product properties, not biological ones. The human hand was never optimised to be produced 100,000 times a year or fixed with a screwdriver, because evolution does not have a supply chain.

This is the distinction worth taking to dinner. When an engineer says a design is better, the honest follow-up is always “better at what, and worse at what?” Rodriguez answered that question himself before anybody asked, which is a good sign about the engineering.

How to Teach Your Kid About Robot Hands

Ages 5–8: Build a tendon, then tape the pinky

Materials: a strip of thick cardboard, three or four drinking-straw segments, string, tape.

Tape the straw segments along the cardboard strip with small gaps between them, thread the string through all the segments, and tape the far end. Pull the string and the strip curls. You have built a tendon-driven finger in about four minutes.

Now the key question: where is the motor? It is your hand, far from the finger. That is exactly how your own fingers work, and exactly what the Atlas designers removed.

Second activity, even simpler. Tape your child’s pinky to the side of their hand with a loop of tape and have them do five normal things: hold a cup, use a spoon, open a door, pick up a coin, write their name. Most of it works fine. Ask which one got hard. Then tell them Boston Dynamics made the same decision on purpose.

Ages 9–12: Run a gripper benchmark

Materials: two clothespins or clips, cardboard, tape, rubber bands, and ten household objects of different shapes.

Build a simple two-finger gripper. Any design that closes is fine. Then line up ten objects: a pen, a coin, a sock, an apple, a glass, a sheet of paper, a key, a ball, a spoon, a pile of rice.

Try each one three times and record success out of three. Add up the score. You now have a dexterity benchmark, and your child has discovered the thing that makes robot hands hard: the objects that defeat a simple gripper are not the heavy ones. They are the thin, flat, soft and granular ones.

Then one modification: add a third finger, or a rubber pad, or a slight concave shape. Re-run the ten objects. Did the score improve? Which objects improved? That comparison is the whole research field in miniature.

Ages 13+: Count the parts, then halve them

This is the actual problem Rodriguez described, scaled down.

Have your teenager build a working gripper from household materials, then write a complete bill of materials: every piece, with a count. Most first attempts come in around fifteen to thirty distinct parts.

Now the assignment: redesign it to do the same job with half the parts. No extra materials allowed. They will discover what Boston Dynamics discovered, which is that fewer, larger components beat many small ones when you have to build the thing repeatedly, and that each part removed also removes a way for the device to fail.

Finish with the arithmetic that makes it real. At 100,000 units a year, saving one part per hand saves 100,000 parts. At ten seconds of assembly time per part, that is roughly 278 hours of labour. The numbers are the point.

The question to ask: “If you could only keep three fingers, which one would you drop, and what would you lose?”

What to Do at Home

Ask “better at what?” every time

This is the single most useful habit in this article and it generalises far past robotics. A claim that something beats a previous design is incomplete until you know the axis. Rodriguez’s own framing, that hands are a ruthless trade-off where you are always giving something up, is the correct default assumption about any engineering decision.

Point out tendons in the body

Your own wrist is a demonstration. So is the back of your hand when you spread your fingers. So is the Achilles tendon. Children who notice that bodies route force from a distance start noticing it in machines too, including bicycle brakes, which are tendon-driven in exactly the same sense.

Treat part count as a design metric

Kids build things. The next time your child finishes a project, ask how many pieces it has and whether it could work with fewer. This is not nagging about tidiness. It is the actual criterion professional engineers use, and it is deeply satisfying once a kid gets it, because fewer parts usually means a cleverer idea.

Separate the demo from the product

A hand that performs beautifully in a video and a hand that can be made 100,000 times a year are different achievements, and the second is harder. The previous Atlas hand, per IEEE Spectrum, performed impressive tasks but was not designed for mass production. Noticing that gap in any product announcement is a durable skill.

What not to do

Do not teach your kid that copying nature is the smart approach. Sometimes it is, and the tendon architecture is genuinely clever. But evolution optimised for different constraints: self-repair, growth, metabolic cost, no factories. An engineer copying biology without asking which constraint applies ends up with a beautiful object that cannot be manufactured. The Atlas hand is the counterexample worth remembering.

What to Watch For Over the Next 3 Months

• Week 4: Look for video or published results of the new hand doing a sustained real task rather than a demonstration, and for any stated figure on cycles before failure. Reliability numbers are the claim that matters and the one companies publish last.
• Month 2 red flags: Coverage that describes the hand as more humanlike. It is less humanlike by design, and reporting that gets this backwards is a signal to check the rest of the article too. Also watch for the pinky coming back, which would mean the trade-off did not hold up.
• Month 3 self-check: Ask your kid to explain why the motors moved into the joints and what that cost. If they mention tendons breaking and fingers getting thicker, they have the trade-off. If they say “because it is better,” ask better at what.

Frequently Asked Questions

Why does the Atlas hand only have three fingers?

Because the motors are inside the joints, and motors take space. Fitting five slim digits is incompatible with direct-drive actuation, so the design kept three fingers and a thumb and gained reliability, serviceability and manufacturability in exchange.

Is this hand better than a human hand?

On manufacturability, repairability, consistency and one specific range of motion, the article argues yes. On overall dexterity, tactile sensing and self-repair, no, and nothing close to it exists. “Better” always needs an axis attached.

What is a back-drivable motor?

A motor that can be turned by an external force rather than only driving in one direction. That property lets the motor act as its own force sensor, because something pushing on the finger shows up as resistance at the motor. It removes the need for separate fragile fingertip sensors.

Why are tendons a problem in robots but fine in bodies?

Bodies repair themselves and replace tissue continuously. A cable in a machine does not. Over thousands of cycles it stretches, loses tension and eventually fails, and re-tensioning is a specialist job. The IEEE Spectrum piece notes the new design eliminates tendons and cables that stretch and break.

What does 13 degrees of freedom mean?

It means 13 independent ways the hand can move, each driven by its own actuator. The previous Atlas hand had 7. For comparison, the Shadow Dexterous Hand has 24 joints with 20 actuated degrees of freedom. More is not automatically better, because every degree of freedom is another thing to control, power and repair.

Will robot hands converge on this design?

Unknown, and worth saying so. One company’s engineering judgement at one point in time is not a settled answer, and Rodriguez’s framing of hands as a ruthless trade-off implies other designers facing different priorities will land elsewhere. Watch what gets built in volume rather than what gets demonstrated.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. IEEE Spectrum. (2026, October 1). “Atlas Robot’s New Hand May Outperform Humanlike Designs.” https://spectrum.ieee.org/robust-robot-hand
2. Wikipedia contributors. “Shadow Hand.” Wikipedia. https://en.wikipedia.org/wiki/Shadow_Hand
3. OpenAI, Andrychowicz, M., Baker, B., Chociej, M., et al. (2018). “Learning Dexterous In-Hand Manipulation.” https://arxiv.org/abs/1808.00177
4. Ackerman, E. (2026, September 15). “Digit 5 May Be the First Humanoid Robot Worker That’s Truly Safe.” IEEE Spectrum. https://spectrum.ieee.org/humanoid-robot-safety
5. Wikipedia contributors. “Moravec’s paradox.” Wikipedia. https://en.wikipedia.org/wiki/Moravec%27s_paradox
6. Wikipedia contributors. “Humanoid robot.” Wikipedia. https://en.wikipedia.org/wiki/Humanoid_robot
7. Occupational Safety and Health Administration. “Robotics.” US Department of Labor. https://www.osha.gov/robotics

Related reading on HiWave Makers: robot grippers, suction and fingers explained, how robot dexterity is measured, and how a humanoid learns laundry through imitation.