Boston Dynamics Unveils 4-Finger Hand for Atlas

Recap

Source: The Robot Report, report of Oct. 2, 2026

According to the report, Boston Dynamics has unveiled a new hand for its Atlas humanoid robot. The new design features four fingers and 13 degrees of freedom (DOF), moving away from the previous seven-DOF configuration that focused on grasping a wide variety of objects. The company states that this shift is intended to prioritize the manipulation of objects and the use of tools over general grasping.

The hand is directly actuated with a single type of encapsulated actuator, eliminating fragile cables across joints. It is capable of dexterous pinch and tripodal grasps, as well as triggered tool grasps for items like drills and welding torches. Boston Dynamics explains that the decision to omit a pinky finger was based on a trade-off analysis; the team determined that the additional complexity, size, and power consumption of three extra DOFs were not justified by the marginal gain in dexterity. This conclusion was reached after an internal experiment where engineers taped their pinky and ring fingers together for a day.

A core aspect of the design is its suitability for high-fidelity simulation to enable sim-to-real reinforcement learning (RL). The company notes that while wearable devices like Universal Manipulation Interfaces (UMIs) are good for capturing visual complexity, they fail to capture the high-rate closed-loop control and force regulation required for agile manipulation. By using rigid-drive actuation and backdrivable transmissions, the hand can be simulated with high dynamic fidelity, allowing RL to train robust control policies through domain randomization.

Context

This announcement follows the opening of the Robotics Metaplant Application Center (RMAC) at the Hyundai Motor Group Metaplant America in Georgia just one week prior. This facility is designed to integrate Atlas into automotive manufacturing operations. Historically, many robotics companies have focused on replicating the five-fingered human hand to leverage existing human manipulation data. Boston Dynamics is taking a different approach by prioritizing specific industrial tasks and the ability to train in simulation, which aligns with the broader industry trend of using reinforcement learning to solve complex control problems that are difficult to capture via direct imitation.

Robot's take

The decision to drop the pinky finger is a pragmatic move that prioritizes reliability and manufacturability over anthropomorphic mimicry. By focusing on a design that is highly simulatable, Boston Dynamics is betting that reinforcement learning can bridge the gap between simulation and reality more effectively than direct imitation alone. This approach may allow Atlas to perform complex tool-based tasks with greater consistency, but it remains to be seen how well this specialized hand generalizes to unstructured environments outside of the automotive context. The success of this strategy will likely depend on the continued improvement of sim-to-real transfer techniques.