Try it online!
There are two types of boppies: owlies and kloppies. The owlies eat the red dots (apples), and the kloppies eat the owlies.
Each creature has a small neural network which tells it what to do. The neural network has two outputs, go forward/backwards and turn left/right. Notice in the gif below the two output nodes. Currently each creature has 5 inputs (+1 bias/threshold), each input shows the distance (0 = no food, 0.5 = very far, 1.0 = very close) to the next food. As in, the neuron fires when it sees food there.
Based on this basic neural network and evolution some interesting behaviour emerges. For example, the prey-predator cycle appears after some time, just as it does in nature:
You can play with it online with a web browser. For better performance you can download the binaries here (Linux, Windows and Mac are all provided, the latter two have not been tested though).
Quickstart: after starting a simulation: press 9 to simulate quickly. After a couple minutes you can press 1 to go to normal speed and inspect what the creatures are now capable of!
H- See help page for up-to-date hotkey list (or click on the question mark in the bottom right corner)W/A/S/D- move camera around (hold shift to move 10x as fast)Mouse click and drag- move camera aroundMouse clickon boppie - follow it until it diesMouse wheel- zoom in/outEscape- stop following boppieO/K- follow fittest owlie/kloppie until it diesP- turn performance mode on/off (off by default). This improves FPS by a lot by turning off all particles and unnecessary visual effects. This is done automatically when changing to a time factor above 32x.F- follow new fittest boppie even after deathC- toggle control of boppie if already following it (i.e. clicked on it before) (useW/A/S/Dto move with it)E- add 5 energy to currently following boppie (press multiple times, then the boppie will grow and have children)-/+- decrease/increase time factor by 2x (min: 0.5x, max: 256x)- Numbers
1through9- set time factor to 2^(number - 1) (e.g.: pressing 9 sets time factor to 256x) Space- pause/unpauseR- show vision rays for all boppiesT- show vision rays for current boppie
- Download the free and open source game-engine: Godot
- Clone this repository:
git clone https://github.com/LiquidFun/BoppieEvolution(or download via github) - Open Godot and import the
project.godotfile - Now press on the play button to run it
- Add obstacles
- Add more senses for the boppies, such as:
- Instead of rays try neural network inputs with information for angle+distance
- Cone for detection of food
- History neuron both input/output, but cap with sigmoid as it easily escalates
- Add sexual reproduction, as in the real world merging DNA from two individuals has greatly benefited survival of the fittest
- Add loading/saving of simulations
- Add areas of high ground productivity (where more food spawns)
- Add seasons where more or less food spawns
- Add (for example), a river in the middle of the map, which separates the species on the left/right of it.
- Encode how many creatures are reproduced in the DNA of the creature (number of children)
- Change meat-eating from boolean to a float, where it essentially becomes meat-tolerance or meat effectiveness (a factor of how much energy can be gained from meat). However high meat-tolerance means low
- Different parallel simulations where the best are merged into one
- Add menu so that simulation could be configured
- Show genetic tree for entire simulation
- Add more graphs:
- Fittest creature each second
- Species stacked bar-plot
- Full screen on neural network, which shows actual weights
- Improve display of recurrent connections
- Use textures on the ground (fertile vs infertile land)
- Show version/commit id in application
- Add eye button to hide eye containers
- Add world configuration tab with seed and other such stuff
- Add high-contrast (bright world background) and large font mode for beamer presentations
- Plots
- Change y-scale once higher values have been achieved
- Ease of use:
- Make the hotkey button more apparent (e.g. show a HOTKEYS text after starting for a couple seconds)
- Add help tooltips for boppie and world window
- Show WASD + SHIFT display when controlling boppie
- Add buttons for common actions
- Eat food, show sensors, stop following
- Load and save simulations
- Alternative display of neural network, where the input neurons are shown at the positions of the sensory neurons near the actual boppie
- Add menu bar at the top: Simulation/Help, with multiple dropdown menu options to restart simulation, back to menu, keyboard help, explanation help etc
- All globals should be in gamecontroller, so that those could be loaded and saved easier
- Neural networks in GDNative C++ instead of GDScript (by far largest bottleneck)
- Separate StatusBar, BoppieBar and WorldStats
- Can the boppies learn recurrent connections if their inputs are delayed
- Instead of sexual reproduction, only reproduce fittest creatures by crossover
- Player vs evolution game mode
- Try to compete with evolution by creating an AI (code in game screen, use eval to evaluate code)
- Get access to better/more sensors over time
- Try to beat ever stronger kloppies
- Currently crossover is not performed when spawning new boppies
- Cannot copy DNA over to other boppies
- Loading and saving does not work
- UI/UX Improvements
- Added progress bars for energy/water/offspring, improving legibility
- Left panel is now for boppies, right panel is for world, status bar has been cleaned up
- Fittest table now in right world panel
- Plants now have plant icon
- Plots
- Current value is now displayed in the top right
- Ease of use:
- Added tooltips for most buttons
- Added buttons for common actions:
- Faster/slower game-speeds, pause play
- Take control of boppie, make invincible, produce offspring
- Performance mode
- Follow fittest boppie
- Show all vision rays
- Added meat after death of boppies
- UI/UX Improvements
- Show additional danger-sense near boppie
- Disabled spike rotation and blood in performance mode
- Improved graphs by drawing lines instead of pixels in texture, graphs can be hidden
- Neurons have colors based on their type, can now drag neurons
- Show importance of neural network connections
- Can now select different profiles for displaying neural network (weights or activations)
- Added lakes, water requirement for boppies and neurons for detecting lakes
- Boppie colors now represent: hue: dna, saturation: energy, luminance: water
- Added reinforcement learning, where neural net weights are adjusted based on rewards
- New neurons/senses
- Terrain resistance sense below and ahead of boppie
- Timer (time encoded in DNA)
- Hunger/thirst
- A single large water ray for detecting water
- Ally sense for detecting allies (same as danger sense)
- Fixed major bug where each time a new boppie was spawned the AIs and NeuronTimers would become orphaned
- Required boppie offspring energy depends on DNA
v0.2.0 (2022-02-17)
- UI/UX Improvements
- Random seed is shown
- Can now drag boppies from fittest list into world
- Camera can now be moved by dragging the mouse
- DNA/Neural network as tabs, as they did not fit in a single column
- Neural networks now show what each input/output neuron means
- Neural networks can have any structure (not necessarily fully connected)
- Added rotating traps with spikes and blood marks after death
- Added these senses for boppies:
- Danger sense (for detecting traps and kloppies)
- Added boppie color as part of DNA
- Implemented NEAT (Neuroevolution of Augmenting Topologies), such that the neural networks of the boppies change with each generation
- Added an innovation number for each connection in the neural network, as described in NEAT
- Crossover of creature DNA and neural networks
v0.1.0 (2021-10-19)
- Two types of creatures (Owlies and Kloppies)
- A basic neural network for each creature
- Creatures have DNA which they pass on to children
- Vision based on 5 rays extending from each boppie
- Evolution simulator by carykh
- Evolv.io by carykh
- MarI/O by SethBling
- Simulating Natural Selection by Primer
- I programmed some creatures. They Evolved. by David Randall Miller
- The Evolution of Predation in a Simulated Ecosystem by The Bibites: Digital Life
- Neuroevolution of Augmenting Topologies by K. Stanley and R. Miikkulainen, 2002