Abstract
Dexterous manipulation is a fundamental capability for robotic systems, yetprogress has been limited by hardware trade-offs between precision,compactness, strength, and affordability. Existing control methods imposecompromises on hand designs and applications. However, learning-basedapproaches present opportunities to rethink these trade-offs, particularly toaddress challenges with tendon-driven actuation and low-cost materials. Thiswork presents RUKA, a tendon-driven humanoid hand that is compact, affordable,and capable. Made from 3D-printed parts and off-the-shelf components, RUKA has5 fingers with 15 underactuated degrees of freedom enabling diverse human-likegrasps. Its tendon-driven actuation allows powerful grasping in a compact,human-sized form factor. To address control challenges, we learnjoint-to-actuator and fingertip-to-actuator models from motion-capture datacollected by the MANUS glove, leveraging the hand's morphological accuracy.Extensive evaluations demonstrate RUKA's superior reachability, durability, andstrength compared to other robotic hands. Teleoperation tasks further showcaseRUKA's dexterous movements. The open-source design and assembly instructions ofRUKA, code, and data are available at https://ruka-hand.github.io/.