Boston Dynamics' new Atlas hand: four fingers instead of five and what it says about choosing a robot for the job

Short answer: according to a The Robot Report article dated October 9, 2026, Boston Dynamics showed the new hand of the Atlas humanoid a week earlier and has now talked about it in more detail. The hand has four fingers and 13 degrees of freedom, that is, independent movements. The pinky has been removed, the gripper has become smaller, and the actuation is direct (directly actuated). The company says it designed the hand for series production and that it can still hold a load of more than 100 pounds, about 45 kilograms. All 13 actuators are identical and enclosed, and no fragile cables cross the joints. The new hand has been used extensively at the headquarters, but it has not yet been rolled out across the whole fleet of robots. What you can take from this interview is not an advertisement for a robot but a list of questions worth asking the supplier of any manipulators. Below are the facts and five questions.

What the company said

All the facts below come from The Robot Report article by Brianna Wessling. It gives the answers of Alberto Rodriguez, director of robot behavior for Atlas at Boston Dynamics. The outlet's site is closed to direct downloading, so I read the article through a page-reading tool and retell it close to the text, without direct quotes.

• What was removed. The pinky has been removed and the gripper made smaller. The company says the hand still copes with a load of more than 100 pounds. Previous hands lifted more than 100 pounds, so extra strength was not needed, and the team considered reducing it slightly for the sake of actuator "transparency." The article does not explain this term; as I understand it, a transparent actuator transmits force with little distortion, which makes it easier to control precisely.
• What is inside. 13 identical actuators, fully enclosed, and not a single fragile cable crossing a joint. Rodriguez links the team's speed of work to the simplicity, modularity and maturity of the actuators: a reliable robot lets the team collect data faster and run more experiments.
• Why they built it themselves. The team studied off-the-shelf hands to understand what the best could do, but built its own because, according to Rodriguez, hands define the value of a humanoid. Every hand requires sacrificing something in favor of something else, and it has to fit the product roadmap.
• What they tested. To set requirements, the team uses a set of representative tasks: handheld power tools, handling cables, fishing out screws. There are also tasks for comparison with earlier hands and for testing the limits, and the team has also tried juggling and speed hand games.
• Why four fingers. The decision was made by intuition, through periods of exploring options, and against target tasks, for example picking up small screws or reaching into a tight space. They looked at several motions: sliding the tip of the thumb along and across the other fingers, rolling it over the index finger in a precise pinch grasp, and a power grasp of a tool handle with a trigger.
• What was rejected. A design with two thumbs, one on each side. It would have allowed the left and right hands to be built on one line, but it added actuation complexity, degrees of freedom, cost and volume.
• How the hand is trained. The actuator is designed for backdrivability and "transparency": as I understand it, this means the hand can be moved from outside without strong resistance, while force is transmitted without large distortions. Residual friction and other non-idealities are compensated by the control software. Thanks to an accurate computer model, the rules of operation that the robot learns in simulation by trial and reward (reinforcement learning) can be transferred directly to the hardware. Rodriguez believes this will make it possible to master operations more complex than "pick and place," especially those that are hard to demonstrate by teleoperation, where contact and pressure change quickly.

What the article does not report: the price of the hand, the timing of series deliveries, the number of units made. I do not name them.

Five questions for the supplier of a manipulator

These are my conclusions from the interview, not Boston Dynamics' words. They suit any robot with a gripper, not only a humanoid.

1. What tasks has the robot done at your place, not in a video? Boston Dynamics named representative tasks: tools, cables, screws. Ask the supplier for a similar list.
   • Who does it: whoever is purchasing.
   • How to check: the supplier has a list of tasks and a result for each, and your task is either on the list or similar to it. If there is no list, mark it as a risk.
2. What happens to the gripper if one actuator breaks? In Atlas, the actuators are identical, enclosed, with no fragile cables.
   • Who does it: an engineer or a contractor whose judgment you trust.
   • How to check: the supplier answers in writing how one actuator is replaced, how long it takes and whether there is a spare.
3. What was sacrificed for simplicity? Here they removed the pinky and rejected the design with a second thumb. Every simplification has a price.
   • Who does it: whoever is purchasing.
   • How to check: ask for a list of what the gripper cannot do and compare it with your operations.
4. How does the robot acquire a skill: by learning from demonstration or in simulation, or by programming the operation? Boston Dynamics is betting on learning in simulation.
   • Who does it: a technical specialist.
   • How to check: ask for a written description of how the robot acquires a skill (training, programming or tuning) and how much time and money it takes to re-adapt it to your part if it differs from those it was prepared on. Confirm this with a trial on your part.
5. Has this been rolled out across the supplier's entire fleet? At Atlas, the new hand has not yet been rolled out across all robots.
   • Who does it: whoever is purchasing.
   • How to check: ask how many robots with this version of the hand are already working and how many hours they have logged.

A calculation with illustrative numbers

Suppose the gripper has 13 identical actuators and the supplier has two spares in stock. One actuator is about 7.7% of all actuators (1 of 13). If replacing one actuator takes 4 hours of downtime and an hour of downtime costs you 5,000 rubles, one breakdown costs 20,000 rubles plus the price of the part. These are illustrative numbers, the source does not give them: substitute your own and ask the supplier for the real failure rate.

When this does not concern you

If your operation is simple and repetitive, the answers to these questions may turn out to be short, but they should not be skipped: the suitability of a gripper depends both on the part and on the working conditions, and the costs and risk of downtime from repairs exist for any equipment. A complex humanoid hand may not be what you need, but you can only find that out by checking against your task.

Summary

According to the article of October 9, 2026, Boston Dynamics kept four fingers and 13 actuators in the Atlas hand, removed the pinky and stated that it is designed for series production; it has not yet been rolled out to all robots. The interview shows what the developers themselves look at: a set of real tasks, the simplicity and modularity of the actuators, the reliability of the robot's work and readiness for training in simulation. These are the same points worth asking any robot supplier about.

I work on AI agents and automation. If you want to look at my projects or discuss your own task, visit my portfolio.

Sources

• The Robot Report, Brianna Wessling, 09.10.2026: Boston Dynamics gives more insight into its redesigned humanoid hand. https://www.therobotreport.com/boston-dynamics-gives-more-insight-into-its-redesigned-humanoid-hand/