Boston Dynamics redesigns Atlas hand for tool use and simulation training

Atlas’s newest hand increases degrees of freedom from seven to 13 while using four fingers. Boston Dynamics has oriented the design toward tool manipulation, mass manufacturing and high-fidelity simulation for reinforcement learning.

For manufacturing operators, manipulating tools and recovering slipping objects could expand the tasks humanoids can perform beyond basic grasping. Whether this translates into dependable factory work remains unestablished by the reported initial results.

Boston Dynamics is shifting Atlas’s hand design from grasping varied objects toward manipulating them, including operating tool triggers and reorienting objects within the hand. The new four-fingered design has 13 degrees of freedom, direct actuation and pressure sensors across its fingertips and palm. The company omitted a pinkie because it judged the additional dexterity insufficient to justify the extra actuators, size and power demand. The hardware also reflects a training decision: Boston Dynamics considers reinforcement learning in simulation essential for fast manipulation and force control, rather than relying on human demonstrations alone. Its actuation design is intended to make simulated behavior transfer more reliably to hardware. The company reports promising initial transfer results, but the article provides no quantified performance or production reliability measures. For manufacturers, the relevant development is a hand designed around tool use and repeatable control—not simply human resemblance.

Evidence supplied by The Robot Report.