Climate robotics / field study

Seeds,
airborne.

An autonomous seed dispersal system for landscapes rebuilding after fire—designed from the sky, tested on the ground.

Explore the work
Drone dispersal test in a field with a pilot and observer
Test site / field deployment
Role Systems / Flight / FDEContext Forest fire recoveryMethod Drone + field researchFocus Regeneration
The premise / 00

Recovery is a coordination problem.

After a forest fire, the land is not an empty canvas. It is a complex system of wind, soil, slope, and surviving life. We asked: could a drone make the first move?

I spoke with farmers, built the dispersal hardware, planned autonomous routes, and tested the handoff between a machine in the air and a landscape below.

Drone carrying a seed dispersal mechanism over a test landscape
Flight test / dispersal rig
Open drone chassis showing electronics and flight hardware
Hardware / flight platform
What I owned / 01—04

The full loop.

Not one isolated feature. A connected system from listening to landing.

01

Seed dispersal mechanism

Designed a gravity-fed release system that metered seeds consistently without compromising flight stability.

02

Flight planning

Mapped burned regions and translated recovery density into equidistant flight paths and drop points.

03

Kinematics + wind

Accounted for drift, speed, altitude, and timing so each release landed where the model intended.

04

Human discovery

Spoke directly with farmers to understand what recovery looks like on the ground—not just on a map.

How the seed dropping works / 02

Inspired by drone fishing.

Drone fishing uses a visual signal: when the drone's LED changes from red to green, a light sensor triggers the release and drops a net into the water. I adapted that same idea for restoration. A color change on the drone became a machine-readable command; the sensor detected it, and the mechanism released a seed at the planned coordinate.

Drone flying over a wetland with a suspended release mechanism
Signal trigger / release test
Mapping the recovery area / 03

The ground tells you what the map can't.

We started with location data, plant data, terrain, and the trees already surviving in each area. Then the route planner found candidate drop coordinates that stayed evenly spaced while avoiding existing plants and unsafe terrain.

The approach is commonly described as a Poisson-disk sampling or constrained coverage algorithm: generate points with a minimum separation distance, then filter them against map boundaries, obstacles, and ecological constraints.

3D forest model showing a dense cluster of trees across a mapped grid
Vegetation model / existing tree constraints