Icebergs make terrible landing pads. The surface is slick, almost never flat, and the wind coming off open Arctic water gusts hard enough to shove a quadcopter sideways on final approach. A team of Canadian roboticists built a drone that lands there anyway. Ice Dart touches down on drifting ice, drives spines into it, and rides along collecting data while the berg travels.

Why fly the sensors out instead of landing a crew

Instrumenting an iceberg usually means putting people onto it from a boat, which is slow and genuinely dangerous. Ice Dart lets the crew stay aboard, fly a sensor package across, and let the ice do the rest. The package logs where the berg drifts and the temperatures it passes through, feeding the drift models that ships rely on. Its first working deployment runs in the Canadian Arctic, gathering data to validate shipboard iceberg-detection systems.

The landing gear is where the engineering went

Two mechanisms carry the whole design. The legs are hinged and fitted with friction brakes instead of springs, so a hard arrival pivots the leg against the brake and dumps the energy into friction rather than launching the airframe back off the ice. Then the feet commit. Each foot carries two needle-like spines, one angled upslope and one downslope, that extend on touchdown and bite in. That pairing holds the drone on faces tilted nearly 60 degrees off level, and the spines retract again when it is time to fly home. Autonomous landing-site selection is the next item on the team's list. Photos of the mechanism are in the original Hackster writeup.

Build the interesting half on your bench

You do not need a boat to prototype spine-anchored landing gear. A single leg on a hinge, a servo, and an accelerometer will teach you most of it:

Two SG90 micro servos (4.8-6 V) as the spine actuators, driven from a separate 5 V rail so the stall current does not brown out your board.
An MPU6050 on I2C at address 0x68, SDA and SCL wired to GPIO 21 and 22 on an ESP32 dev board, with 4.7k pull-ups if your breakout does not carry them.
A hinged leg cut from 3 mm acrylic with a felt-and-bolt friction pivot, so you can feel the difference between a bouncing leg and a braked one.
A 3.7 V LiPo and a foam-block test rig on a tilting board you can prop at 20, 40, and 60 degrees.

The gotcha is timing. Fire the spines off the Z-axis acceleration spike, not a fixed delay after the descent command, because a timer will drive them before the foot is loaded and they will skid across the surface instead of digging in. Log the raw accelerometer stream to serial for a few drops first, then set the threshold from what you actually measured.

Originally published on blog.circuit.rocks.