Quasi-direct-drive actuator

What it means

A quasi-direct-drive actuator combines a high-torque-density motor with a relatively low-ratio transmission. It retains a reduction stage, unlike direct drive, but aims to preserve useful backdrivability and control bandwidth. Yu and colleagues use this definition in their hip-exoskeleton actuator study. The term describes an actuation approach rather than a single gearbox product or a guaranteed performance level. [1]

The mechanical tradeoff matters because reduction increases the rotor inertia felt at the output. Modern Robotics explains the square-law relationship between reduction ratio and apparent rotor inertia. Choosing a modest ratio can therefore change the behavior of an externally loaded joint, alongside friction and other actuator properties. [4]

How it applies to humanoids

Humanoid designers need both locomotion actuators and manipulation actuators. Quasi-direct drive has been investigated in related platforms serving both areas. Gealy and colleagues use it in Blue, a human-scale robot arm for force-controlled manipulation, including a VR interface for telepresence and demonstration collection. That is arm research, not evidence that a complete humanoid achieves the same results. [2]

Benjamin Katz's MIT thesis develops a modular actuator for highly dynamic robots. It describes applications in a quadruped, a lower-body biped and a bilateral teleoperation and haptic-feedback system. These examples show the breadth of the engineering problem without making the actuator a general measure of humanoid capability. [3]

1X's July 2026 update explicitly identifies NEO's hand actuation as quasi-direct-drive tendons. That attribution concerns the documented hand assembly. [5]

Terminology and comparison traps

The cited QDD papers describe particular designs and experiments, not a universal certification threshold for the name. Before comparing two actuators, ask for their actual reduction ratios, output torque, mechanical resistance and control measurements. A label alone cannot establish that their behavior is equivalent. [1, 2]

Bandwidth and torque need context. The hip study compares torque capability, tracking accuracy and backdrivability, while the Blue work reports position-control behavior. These measurements answer different questions and should not be combined into one actuator ranking. [1, 2]

Low cost is also a design-specific claim. A research arm's proposed cost does not establish a humanoid's purchase price, manufacturing cost or service expenses. Read which system and production assumptions the cited figure describes. [2]

How to read it on humanoidversus.com

Use actuation as the broad energy category and max_joint_torque_nm for the documented peak of a named joint. Neither field identifies a QDD transmission. The robot's technical sources must explicitly describe its mechanism before the label is attributed.

NEO is linked because its record cites the official hand update naming quasi-direct-drive tendon actuation. This entry has no live table. Treat the stated mechanism as evidence about that hand configuration, and check separate technical sources before applying the label to other joints or generations. [5]

Further reading

Yu and colleagues study a wearable actuator, Gealy and colleagues study manipulation, and Katz's thesis discusses dynamic robot applications. Modern Robotics supplies gearing background, while 1X documents a humanoid hand implementation. [1–5]