STRANDBEEST WALKING ROBOT

FIG. 01One crank drives every joint. The link lengths shape the foot path so its bottom is nearly flat, which keeps the body level while the foot is planted, then lifts the foot clear on the return. Each leg runs a third of a turn behind the next so some feet are always on the ground. Move the pointer across the figure to turn the crank by hand.

Theo Jansen’s Strandbeests are wind-powered kinetic sculptures that walk on beaches. Sails catch the wind, and that energy turns a crank that drives his Jansen linkage legs, which turn rotation into a surprisingly lifelike walking gait. In Spring 2023, a six-person team of us at Texas A&M TURTLE Robotics set out to build a motor-driven version as a fun showcase robot for the club.

The assembled Strandbeest prototype

The Linkage

The whole robot depends on getting the leg geometry right. A Jansen leg is a single degree of freedom mechanism: one crank rotates and every other joint follows, tracing a foot path that is flat along the bottom (the stance phase, when the foot pushes on the ground) and lifts clear on the return (the swing phase).

Our initial research settled three things before any CAD:

  • the ideal leg linkage dimensions, from the existing research on Jansen’s linkage,
  • the leg configuration, meaning how many legs to run and how to phase them around the crank so some feet are always on the ground, and
  • the motor requirements: low speed and high torque, since the crank turns slowly and has to lift the whole body.

Jansen linkage joint layout for a mirrored leg pair

The foot path traced by one leg as the crank turns

Design and Fabrication

We modeled everything in SolidWorks, scaling the research dimensions into real parts. The final assembly has 8 leg assemblies driven through 4 gear assemblies.

  • Gears and leg linkages: laser cut.
  • Chassis: a base plate cut on a scroll saw and painted, with laser-cut side plates.
  • Hardware: 6 mm steel axles running in 6 mm ID bearings, with a variety of spacers to keep each linkage layer in its own plane.

SolidWorks model of the full robot

Control System

The electronics are deliberately simple:

  • a 22.2 V, 5200 mAh battery feeds an ODrive 3.6, with a brake resistor to absorb regenerated energy, driving two brushless motors,
  • a CUI AMT10 encoder (8192 CPR) gives the ODrive closed-loop feedback,
  • a 5 V, 3 A step-down powers a Raspberry Pi 4B running Debian Bullseye, and
  • a small travel router puts the Pi on its own network.

The Pi talks to a pygame dashboard over WebSockets, so the robot can be driven and monitored from a laptop.

Power and data diagram

What We Would Change

At the end of the semester, our list of changes for the next iteration was:

  • Increase joint stability. Each leg has many pinned joints, and play in each one adds up at the foot.
  • Add feet so it can walk on softer or uneven surfaces.
  • Add ribs across the base plate for stiffness, and reduce the weight of the main chassis plate.
  • Add sensors and autonomous operation. This version is driven by hand from the dashboard.

For me, the project was a lesson in how much a mechanism’s behavior lives in its geometry. The motors only turn cranks, so the gait is decided by link lengths and phasing in CAD long before any control code runs, and most of what we wanted to fix afterward was mechanical: stiffness, tolerances and weight.