Boston Dynamics Deleted a Finger to Build a Better Hand

Boston Dynamics gave Atlas a more dexterous hand by adding six degrees of freedom and removing a finger.

That sounds contradictory only if a robot hand is judged by how closely it resembles a human one. Boston Dynamics is judging the mechanism by the work it can do, the data it can learn from, and the complexity it has to carry into manufacturing.

The new Atlas hand has four fingers and 13 degrees of freedom, up from seven in the previous generation. Its joints are directly actuated with a single actuator type, and dense pressure sensing covers the fingertips and palm. Boston Dynamics says the design shifts the goal from grasping a broad set of objects toward manipulating them.

A hand is a robot inside the robot

Every additional controllable joint adds more than motion. It adds an actuator, sensing, wiring, control, calibration, simulation complexity, power demand, volume, manufacturing work and another component that can fail.

The new hand gives the thumb four degrees of freedom and the other fingers three each. Boston Dynamics says that geometry supports finger splay, fingertip control, pinch and tripod grasps, in-hand reorientation, recovery from slipping objects and tool use while operating triggers.

Those capabilities attack a bottleneck Cyberdelia has been tracking across humanoid robotics: impressive locomotion does not automatically produce useful manipulation. A robot can cross a room and still fail at the ordinary hand work that makes human environments valuable.

The missing pinky is the engineering story

Boston Dynamics says the team debated a fifth finger and even experimented by taping their own pinky and ring fingers together. Their conclusion was that the extra capability did not justify three additional degrees of freedom.

A fifth finger would mean three more actuators, with additional cost, volume, power consumption and failure probability. The team kept four.

This is not minimalism for its own sake. It is a design rule: a part has to buy enough capability to justify every burden it adds to the system.

Human similarity still matters, but as an interface

Boston Dynamics also wants to reduce what roboticists call the cross-embodiment gap. Human demonstrations are abundant compared with robot demonstrations. A hand sufficiently similar in scale and kinematics can make human manipulation data easier to translate into robot learning without requiring the mechanism to be an anatomical copy.

The result is a compromise. Atlas needs to fit into spaces and use tools designed around people, but the hand also has to be simulated cleanly, manufactured repeatedly and repaired at industrial scale.

That distinction connects directly to China Can Mass-Produce Humanoids. Can It Mass-Produce Competence?. Hardware scale and behavioral competence are separate variables. The hand sits exactly where they meet: more dexterity creates richer behavior, but uncontrolled mechanical complexity makes fleets harder to build and maintain.

Tactile sensing closes another loop

Dense pressure sensors on the fingertips and palm give the controller information that vision alone cannot provide. Contact pressure can reveal a slipping grasp, a small collision or the onset of motion before an object visibly falls.

Boston Dynamics' public material demonstrates the architecture and intended skills. It does not yet establish fleet-scale reliability, cost per useful task-hour, or how much of the demonstrated manipulation transfers to unfamiliar objects and environments.

Those are the measurements that matter next. The four-finger hand is not interesting because it looks futuristic. It is interesting because its designers are explicitly trading theoretical capability against the brutal accounting required to ship a physical machine.