A C++ / ImGui simulation of Thomas D. Seeley's Honeybee Democracy: scout bees waggle-dance to advertise candidate nest sites, and a genetic algorithm searches for the strategy that lets a swarm reach the best consensus.
Every spring a honeybee colony grows too big for its hive and roughly 10,000 workers leave with the old queen to find a new home. The swarm clusters on a nearby branch while a few hundred scout bees fan out, inspect tree cavities, and come back to report. Each scout that returns dances on the swarm's surface to advertise the site she found.
The dance is a tight figure-eight whose angle from vertical encodes the direction to the site relative to the sun, whose duration encodes the distance, and whose vigor and number of repeats encode the cavity's quality — a great site gets a long, enthusiastic dance. Other scouts watch, fly out to evaluate independently, and either return enthusiastic or not. Crucially, every dancer's enthusiasm decays over time, which keeps an early scout's stale assessment from dominating. Once enough scouts coincide at one site, the swarm commits and takes off together.
This decentralized recruit–evaluate–decay–commit loop finds the best of dozens of candidate cavities almost every time. Tom Seeley spent decades reverse-engineering it and lays the full story out in Honeybee Democracy (2010), including its parallels to good human decision-making.
The simulation is a simplified, headless version of that scout-and-dance loop, parallelised across many small swarms so a genetic algorithm can search for the best strategy. Each Hive holds ~100 scout bees and lives at a fixed location in a shared 2D world. Around them, ~20 candidate nest boxes are scattered in [-1, 1]²; their qualities are sampled from u⁴ so most boxes are poor and one rare box is ideal — the swarm has to find that needle.
Each scout runs a five-state machine — Rest → Search → BackToHome → Dance → Rest or Rest → Follow → Find → BackToHome → Dance → Rest — controlled by four real-valued strategy genes per swarm:
| Gene | What it controls |
|---|---|
random_chance |
Probability per step that a resting bee leaves to scout a new site at random |
follow_chance |
Probability per step that a resting bee follows a dancing nestmate |
linear_decay |
How fast a dancing bee's enthusiasm bleeds away |
dance_force_exponent |
How strongly site quality maps to dance vigor |
Several swarms with different gene values share the world and run the same exploration task for thousands of steps; a swarm's fitness is the maximum consensus_fraction × site_goodness it ever reaches during a run. Between generations the underperforming swarms are crossed with the best swarm and jittered, and every 15th generation the bottom 10% are wiped and re-seeded with random genes. Over many generations the algorithm converges on swarms whose random / follow / decay / dance-vigor balance reliably steers their bees to the rare good nest.
You need a C++17 compiler, CMake ≥ 3.14, Git, and OpenGL + windowing system development headers. Everything else (ImGui, ImPlot, GLFW, Eigen) is fetched and built by CMake — you don't have to install them yourself.
Tested on Linux and macOS; Windows is untested but should work.
Linux — Ubuntu / Debian
sudo apt update
sudo apt install build-essential cmake git \
libgl1-mesa-dev libxinerama-dev libxcursor-dev libxi-dev libxrandr-devLinux — Arch / Manjaro
sudo pacman -S --needed base-devel cmake git \
mesa libxinerama libxcursor libxi libxrandrLinux — Fedora
sudo dnf install gcc-c++ cmake git \
mesa-libGL-devel libXinerama-devel libXcursor-devel libXi-devel libXrandr-develMacOS
xcode-select --install # if you don't already have the Xcode command-line tools
brew install cmake git # if you don't already have Homebrew, install it from https://brew.shWindows
Install Visual Studio 2019 or newer (the free Community edition is fine — pick the "Desktop development with C++" workload), plus CMake and Git for Windows. Run the build below from a Developer Command Prompt for VS.
git clone https://github.com/brenocq/honeybee-democracy.git
cd honeybee-democracy./build.sh -x # release build, then runThat's it — a window should pop up with the simulation. Press Start to begin.
Other useful flags (./build.sh --help lists them all):
./build.sh -h # show help message
./build.sh -d -x # debug build, then run
./build.sh -j 4 # limit to 4 parallel compile jobs
./build.sh # just build, don't runThe first build takes a few minutes because CMake fetches and compiles GLFW, ImGui, ImPlot and Eigen from source. Subsequent builds are incremental and finish in seconds.
The Simulation window has three stacked plots driven by a single set of controls — Start / Pause / Stop / Reset plus a Real-time checkbox that paces the simulation at one step every 25 ms so you can actually watch the bees fly.
Environment plot — the 2D world. Hives appear as colored squares, their bees as small dots in the same color, and nest boxes as diamonds coloured by quality on a Jet colormap (blue = poor, red = ideal).
Fitness plot — one line per hive showing its best-of-run fitness over generations. The lines climb and converge as the GA finds better strategies.
Consensus plot — one mini polar dance per hive, side by side. For each nest box that some of a hive's bees are currently dancing for, a rectangle radiates from the hive, its angle pointing toward the box, its length matching the distance to the box, and its width proportional to the fraction of bees dancing for it. Watch a hive narrow from many short rectangles to one long bold one as consensus forms.