Boston Dynamics Redesigns Atlas Hand for Scale Over Human-Like Looks

Boston Dynamics today announced a redesigned hand for its Atlas robot, shifting from research hardware to a product intended for scalable manufacturing. The previous generation used three fingers to perform remarkable, superhuman feats, but it was never designed for production in tens or hundreds of thousands.

That shift makes once-secondary questions central: how to build a hand that can do everything required while remaining rugged, reliable, and cost-effective. The new Atlas hand does not closely resemble a human hand, but it may be the version that reaches the real world.

Compared with the eerily human-like hands from many other humanoid robotics companies, Atlas's new hand looks clunky. Those more anthropomorphic hands can perform dexterous manipulation that is incredible, as longtime robot observers might say. They also look like slim, graceful mechanical copies of human hands, seeming like science fiction made real.

Alberto Rodriguez, Director of Robot Behavior at Boston Dynamics, tells IEEE Spectrum that hands are a ruthless design tradeoff. He says there is no way around it: you are always giving up on something, and many current designs give up on reliability and manufacturability.

The unavoidable result of cramming all necessary functions into a human hand form factor is a hand that is either fragile or very expensive, and likely both. For now and the immediate future, such hands are mostly suited to research and demos.

With the new Atlas hand, Boston Dynamics is making design tradeoffs to achieve a hand that, according to the press release, can handle a vast array of useful tasks including tool use; is strong and rugged; can be cleanly simulated; can be mass manufactured and repaired at reliably low cost; and is close enough in form factor to a human hand that human demonstrations can train it.

Noticeably absent is any requirement that it look cosmetically like a human hand, and it does not. Rodriguez explains that removing the pinky was a straightforward decision after his team spent a day with their pinkies taped to their ring fingers. But why add a fourth finger at all, when the previous Atlas hand seemed to do well with three?

The change is not just an extra finger but a complete redesign that nearly doubled the degrees of freedom, from 7 to 13.

Rodriguez says Boston Dynamics also added the ability for the fingers to splay open, because it knew that one finger needed to move against another and that the hand needed to hold a tool handle with a trigger. This includes common tools such as drills or welding torches.

Rodriguez says he is convinced that from a product perspective, this hand should be able to do everything needed for the foreseeable horizon. The splay ability goes beyond what human hands can do. According to Rodriguez, reinforcement learning can discover uses for these superhuman extra motions and exploit them.

Reliability, manufacturability, and cost are closely related. By reducing degrees of freedom to just 13, relative to more anthropomorphic designs, Boston Dynamics can use fewer, larger, and more powerful actuators. The actuators are embedded directly in the joints in a direct-drive configuration, with a transmission that lets the motors be back-driven and react to force and contact.

Each actuator pack is a single unit that can be easily replaced, and there are no delicate tendons or cables across joints to stretch and break. Rodriguez says the company is currently working on figuring out what needs to change in the fine details of the design if it wants to make 100,000 of these hands a year.

Not every humanoid company will make the same tradeoffs, and there remains plenty of room for creativity in hand design. But if humanoids with hands are going to scale, difficult decisions will have to be made, and compromise will be necessary to deliver on the promises the humanoid industry has committed to.