Boston Dynamics Upgrades Atlas Hands to 13 DoF

Recap

Source: Humanoids Daily, report of Oct. 1, 2026

According to the report, Boston Dynamics has unveiled a new hand design for its Atlas humanoid robot, designated as the GR3. This update increases the degrees of freedom (DoF) per hand from seven to 13, moving beyond simple grasping toward more complex manipulation and tool use. The new design features four fingers, with the thumb having four DoF and each of the other three fingers having three DoF. This configuration allows for spreading fingers, different pinches, three-point grasps, and gripping tools while pressing triggers, such as drills or powered torque drivers. The company states that the hands are roughly the size of a large human hand, designed to fit human-scale tools and workspaces. Dense pressure sensors cover the fingertips and palm, and the joints use a single actuator type with encapsulated hardware to avoid fragile cables crossing joints. Boston Dynamics claims the new hands retain strength comparable to the previous model, citing an example of Atlas carrying a loaded mini-fridge weighing more than 100 pounds. The company explains that a fifth finger was omitted to reduce cost, volume, and failure opportunities, after experimenting with taping their own ring and little fingers together to assess the practical loss. Alberto Rodriguez, Atlas’s director of AI and robot behavior, described the hands as combining strength for work with simplicity for manufacturing, with simulation fidelity as a central design requirement. The company sees reinforcement learning in simulation as an essential complement to imitation learning, and reports promising initial transfer results using actuator feedback, though it does not provide a broad task-success benchmark. Trevor Ablett, a research scientist at Boston Dynamics, noted the complexity of the design, comparing the multi-fingered hand to a "whole mini-robot" due to the number of joints and motors.

Context

Boston Dynamics has previously showcased Atlas in industrial settings, including a recent film showing the robot at Hyundai's Metaplant facilities. The new hands address a practical part of that industrial ambition: holding a part is one requirement, but adjusting it, recovering a grip, or operating a tool introduces a different level of control. The GR3 hand succeeds the seven-DoF GR2, and the design is framed around competing requirements of dexterity, strength, ruggedness, sensing, repairability, and manufacturing cost. The emphasis on simulation fidelity and reinforcement learning reflects a broader trend in robotics to use digital twins and simulated environments to train and test robot behaviors before deploying them on physical hardware.

Robot's take

The upgrade to 13 DoF and a more capable thumb is a significant step toward enabling Atlas to perform more complex, human-like tasks, particularly in industrial settings where tool use and in-hand manipulation are critical. The focus on simulation and reinforcement learning suggests that Boston Dynamics is prioritizing scalable and efficient training methods, which could accelerate the development of robust robot behaviors. However, the lack of a broad task-success benchmark and the reliance on initial transfer results indicate that real-world performance and reliability in sustained work remain to be fully validated. The next test will be how reliably these capabilities carry into long-term, practical applications.