Website Repository

Lakehopper is my vision for an autonomous drone that can fly long distances by hopping from one lake to the next, recharging its batteries each time using solar panels.

As my Master’s thesis, I developed Lakehopper’s high-level planning software. This system uses a convolutional neural network to identify lakes and buildings from aerial imagery. From this, it generates a navigation graph to calculate the best multi-hop paths between lakes. These paths avoid build-up areas and restricted airspace. This software uses Python and Tensorflow 2 to create and train the models. The planner component that generates the navigation graph and calculates shortest paths is written in Rust and also acts as the server for a pure-TypeScript web interface.

In my free time I’m working on the hardware of the drone. The first version I developed unfortunately experienced a crash rapid unscheduled disassembly on its maiden flight. I’m currently working on the second version.

Predictions made by Lakehopper's vision CNN for lake and building
Segmentation model predictions (GT: Ground Truth, ENB3: EfficientNetB3, MNV2: MobileNetV2)

Screenshot of a map with a multi-hop path between lakes planned by
Lakehopper's planning component
Planner path with four hops between lakes

Lakehopper 1 on a workbench
Lakehopper 1 on a workbench

Screenshot of Lakehopper 2's design in Fusion 360 CAD
Preview of Lakehopper 2's design

Screenshot of the browser UI to control Lakehopper's planner
Browser UI to the planner software