Two or three joints per leg: degrees of freedom and what they cost in a toy quadruped

The first question in a desktop quadruped project is not “how fast can it run” — it is “how many joints per leg”. The answer decides the actuator count, the cost structure, the battery budget and the list of things the robot will never do. It also decides the shape of everything downstream, from the gear train to the app’s animation set.

Two DOF: the honest walker

Two joints per leg — hip and knee, in one plane — can produce a walk, a trot and a turn that convince on flat floors. What they cannot do is change body height, lean, or handle uneven ground gracefully. For a desktop companion product, that is often enough: the robot lives on a desk or a hard floor, and its appeal is character rather than terrain. Two-DOF legs halve the actuator count against three-DOF designs, which shows up in four places at once: cost, weight, battery and the reliability math across sixteen joints instead of twenty-four.

Three DOF: articulation has a price

Three joints per leg add lateral hip motion, which brings height changes, body leans, better stability on uneven floors and the crouching poses that make videos delightful. The price is paid four times over: twelve actuators instead of eight, more gear trains to keep quiet, more joints to wear, and a control problem that grows non-linearly. Three-DOF legs make sense when body articulation is central to the character — when the robot needs to sit, crouch and lean as part of its expression. When walking alone carries the product, three DOF is cost without payoff.

The middle paths

Between the two sit the compromises worth knowing: three-DOF on the front legs and two on the rear; or two-DOF legs with a central spine actuator that raises the whole body. Both capture much of the articulation appeal at a fraction of the actuator count. The spine trick in particular — one actuator lifting all four legs together — delivers the height-change wow in videos at one quarter of the joint cost.

Gaits are tuned for life, not for demo

Demo videos run gaits near actuator limits: fast, dramatic, battery-emptying. Shipping products tune the same gaits slower and gentler, because the retail robot walks for years, not for takes. The tuning work is real: foot contact detection, speed limits by surface, and a stall-and-recover behaviour for when a foot catches a rug. A quadruped that walks at half its demo speed but walks every day outlives the one that performs for the camera and limps in the living room.

We develop desktop quadrupeds from leg architecture up — joints, gaits, gear trains and the behaviour model above them. The OEM/ODM page describes how we run development, or write to us with the motion repertoire you want your robot to have.