Hardware & Sensing experiments · under construction
🐝 BEE-NAV
Before a honeybee forages, it flies a short loop around the hive, learning what home looks like. TU Delft turned that trick into a navigation system: a neural network of just 3.4 to 42 kilobytes that learns a location from one short learning flight and runs on a Raspberry Pi 4. Their code is open (MIT). BEE-NAV is my workbench for reproducing their experiments and, eventually, flying it on cheap hardware.
Data
- tudelft/Bee-Nav code (MIT)
- Omnidirectional images from local test flights (later)
Roadmap
- Clone and run TU Delft's simulation
- Reproduce the path-integration noise experiment
- Hardware bill of materials
- Retrain on a local test area