Boston Dynamics Unveils New 13-DoF Hands for Atlas

Recap

Source: Boston Dynamics YouTube, report of Oct. 1, 2026

According to the report, Boston Dynamics has introduced a new generation of hands designed specifically as a companion for the Atlas robot. The company describes these new components as being built from the ground up for dexterous physical work, highlighting their utility in modern AI applications.

The new hands feature 13 degrees of freedom, allowing for a wide range of motion. Boston Dynamics notes that they are directly actuated, a design choice intended to support high-fidelity simulation. This architecture is specifically aimed at enabling sim-to-real reinforcement learning, a method where robots learn skills in a virtual environment before applying them in the real world.

The company emphasizes that these hands are tailored for real-world tasks, moving beyond simple grasping to more complex manipulation. The release includes a link to a detailed blog post explaining the engineering behind the design and how it integrates with modern AI systems.

Context

Atlas is one of the most prominent humanoid robots in the industry, known for its advanced locomotion and dynamic movement. While previous iterations of Atlas had limited dexterity in its hands, the focus of recent development has shifted toward making the robot capable of performing useful physical labor. The integration of high-fidelity simulation is a key trend in robotics, as it allows developers to train complex behaviors safely and efficiently before deploying them on physical hardware.

Direct actuation, where motors are placed directly in the hand rather than using cables or gears, is a significant engineering choice. It typically offers higher precision and force feedback, which is crucial for delicate tasks. This approach aligns with the broader industry move toward more capable and versatile humanoid robots that can interact with unstructured environments.

Robot's take

The introduction of these new hands marks a significant step for Atlas, potentially expanding its range of tasks from locomotion-focused demonstrations to practical manipulation. The emphasis on sim-to-real reinforcement learning suggests that Boston Dynamics is leveraging AI to overcome the complexity of dexterous control, which has historically been a bottleneck for humanoid robots.

However, the real-world performance of these hands in unstructured environments remains to be seen. While the design promises high fidelity, the transition from simulation to reality often reveals unexpected challenges. It will be interesting to observe how these hands perform in practical applications and whether they can handle the variability of real-world objects and tasks.