Walking AT-ST
A scale bipedal walker taken from LEGO mechanism study to Fusion 360 assembly to 3D-printed servo hardware — including a hand-derived inverse kinematics solution for the legs.
The problem
A two-legged walker is a hard mechanism. It has to carry its own weight, stay balanced through a gait, and do it with linkages and actuators that fit inside a shape nobody designed with statics in mind. The AT-ST’s silhouette is all overhanging body and thin, reverse-jointed legs — which is exactly the wrong mass distribution for walking, and exactly why it’s an interesting thing to try to build.
This has been my long-running project since 2023, and it’s where most of what I know about mechanisms and electronics actually came from.
Prototyping the mechanism in LEGO
I didn’t start in CAD. Before committing to printed geometry I built the leg mechanism in LEGO Technic driven by an EV3 controller, because it let me change link lengths and joint positions in minutes instead of hours.
The point wasn’t to make LEGO walk well. It was to find out which linkage arrangement produced a usable foot path before any of it was expensive to change.
The kinematics
Once the leg had a defined geometry, the question became how to actually command it. Each leg is a two-link chain, so driving the foot to a point means solving backwards from the target position to the two joint angles.
I worked the inverse kinematics out by hand — law of cosines for the knee angle, then an arctangent for the hip, using the link lengths straight off the CAD model.
That sheet is the bridge between the mechanical design and the code. Without it the servos are just three arbitrary angles; with it, the foot goes where you tell it.
CAD
With the geometry settled, the whole thing was modelled as an assembly in Fusion 360 — body, hips, the reverse-jointed legs, and the clawed feet.
Printed hardware
Printed parts, micro servos at each joint, and a controller board mounted on top of the body.
Comparing the printed part against the model on screen is where the iteration happens — tolerances at the joints, clearance for the servo horns, whether a link is stiff enough at the printed wall thickness.
The feet were their own problem: they carry the whole load at the moment of contact, and on this design they’re also the most visually distinctive part.
Result
Still ongoing, which is the point of it. It’s the project where I get to be wrong cheaply — and the loop of prototype the mechanism, derive the math, model it, print it, find out what I missed is the closest thing I’ve had to real engineering practice outside of a lab.