Natural Selection simulator

This project is a 2D simulation that investigates the process of natural selection, a fundamental principle in the theory of evolution originally proposed by Charles Darwin. In this simulation, both predators and prey compete for survival. Each entity possesses an artificial neural network as their "brain" and uses their field of view as input to navigate the environment. At each generation, a random mutation appears, which can prove advantageous or disadvantageous to their survival. The ultimate goal is to survive long enough to reproduce, and entities that develop favorable mutations will be more likely to do so.
This simulation offers a compelling illustration of the fundamental principles that underlie natural selection, providing insight into how organisms adapt and evolve in response to their environment.

Simulation Mechanics

At the start of the simulation, a small number of entities are spawned, comprising both predators and prey. The maximum population of prey is kept much higher than that of predators to prevent their large-scale extinction. Both predators and prey share the following properties:

• Speed and an angular velocity to move around the map
• Moving consumes energy
• A split charge. When full, they split and generate a new entity with a mutation. The charge must then be replenished.

Preys

Prey are represented as green dots. The field of view of a prey, represented as a lidar, is shown in the image below.

The properties of prey are as follows:

• A wide yet short field of view
• The split charge restores over time. Once it has reached a certain period, they split
• When their energy drops to 0, they must rest and not move until they have reached a minimum level of energy
• They can also move backwards

Predators

Predators are represented as red dots, and their field of view is shown in the next image.

The properties of predators are as follows:

• A narrow yet long field of view
• The ability to eat prey, recover energy, and fill their split charge
• The split charge empties over time
• When their energy drops to 0, they die
• After eating, they have a digestion period during which they cannot restore energy or split charge. This is to prevent massive and pointless over-reproduction of predators when hunting large prey colonies
• They can only move forward

Neural network

All entities possess an artificial neural network as their brain. The inputs to the network are the lidar rays of the field of view, and the outputs are the entity's speed and angular velocity. The neural network has weights (represented in orange if w > 0), biases (the neuron is filled if b > 0), and activation functions (tanh).

Predators can only detect preys in their field of view, and vice versa. The closer the prey, the higher the input.

Mutations

Each time a new entity is born, it inherits the same neural network architecture as its parent, as well as a random mutation, which can manifest in the following ways:

• A modification of weights
• A modification of biases
• A new connection (a path of weights)

The process of evolution

At the beggining of the simulation, entities have a very primitive brain, and only a few entities acquire the necessary mutation to facilitate movement. Those predators that remain stationary eventually die, while those that move increase their likelihood of prey capture and reproduction. The preys do not require movement at this stage, and hence, the absence of the mutation for movement neither proves favorable nor unfavorable to their survival.

After a short time, a phenomenon begins to appear: emergence. Emergence happens when complex systems or patterns arise from simple interactions between individuals. In this case, the predators begin to move in colonies or groups, enhancing their survival chances as a collective, rather than as individual predators. This grouping enables the predators to cover a larger area in search of prey.

At this point, only predators that move remain, excluding a few that may have lost their ancestor's mutation that facilitated movement due to a mutation. These stationary predators will ultimately die. As the predator population grows, the prey population significantly declines. The predators will face a shortage of food, which subsequently decreases their population.

Now, this is interesting times. The only preys that survive are those with exceptional speed, agility, or reflexes, while the surviving predators are those with the capability to pursue and capture prey. The remaining few predators that are not skilled hunters, but only lucky, will also survive. The situation is a crisis that both species must survive to ensure their continued existence. The survivors will reproduce and propagate new generations that inherit these useful skills.

The process repeats, with the better-adapted individuals of both species surviving, the preys reproducing rapidly, the predators having access to more abundant food and also reproducing, the prey population declining, and the predator population decreasing too. Subsequently, both species will face another crisis that will enhance their skills. These cyclic crises are inevitable in this environment and do not always result in a positive outcome. Sometimes, predators are unable to hunt preys and become extinct, while other times, preys are all hunted during predator overpopulation, leading to the extinction of both species.

Ovservations

Here are some observations made during the simulation:

1. The predator, framed in blue, has developed the appropriate neural connections to chase and capture preys, giving it a significant advantage over other predators and enabling it to reproduce at a faster rate.
   

2. Some preys do not require exceptional speed. For instance, one prey has acquired a skill that enables it to dodge predators. However, this skill may prove futile if a predator colony surrounds it.
   

3. Some predators have adopted a waiting strategy. They remain stationary until prey enters their field of view, allowing them to quickly hunt the prey. This approach presents a vital advantage since the predator no longer loses energy moving, and thus can survive for longer.
   

Conclusion

Survival in this environment is a daunting task, and in most simulations, the entities do not overcome the second crisis. The constant parameters such as energy depletion, maximum speed, and field of view range have been meticulously adjusted to enhance survival. However, after numerous simulations, the pattern described above remains consistent. The best-adapted individuals that can cope with the prevailing conditions tend to survive longer.

"It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is the most adaptable to change." - Charles Darwin

How to run the simulator

The project has been done in 2 languages: C++ for running and building the simulation, and Python for visualizing it in a vide-like format. You can visualize an already run simulation, run a new simulation with fixed parameters, or adjust the parameters, build and run the simulation.

Visualize an already run simulation

Requirements

Apart from Python 3, the following python libraries are required to be installed:

• numpy (1.19.5)
• pygame (2.1.2)

Steps

1. Go to python_files/ directory.
2. Run the natural_selection_interface.py program with python.
3. A pygame window, named "Simulation", should open. Here you can visualize the simulation.

Run a new simulation with fixed parameters

Requirements

• The same as in "Visualize an already run simulation".

Steps

1. Run Natural-Selection-Proj.exe to run a new simulation. This might take a few minutes, depending on the computer.
2. The simulation will be automatically saved in simulations/ in 2 files: simulation1.bin and neural_nets1.bin. So open python_files/natural_selection_interface.py to edit the code and change the file_num to "1" in line 234.
3. Run natural_selection_interface.py again to visualize the new simulation.

Adjust the parameters, build and run a new simulation

Requirements

• C++ installed, as well as the compiler.
• An IDE (such as Visual Studio) to edit the code, build and compile the project.
• The same as in "Visualize an already run simulation".

Steps

1. Open the repository folder with the IDE and access the cpp_files/.
2. Edit the parameters you want. These are found in entity.h, prey.h, predator.h, neural_network.h and main.cpp. In the header files, the parameters are the ones that are static in the class declaration.
3. You can optionally change the file_num parameter in main.cpp and in python_files/natural_selection_interface.py, so that the simulation is saved in another file instead of replacing the old ones. You will also have to create a new simulation<x>.bin and neural_nets<x>.bin in the simulations/ folder. For example, if file_num = 2, then name the files simulation2.bin and neural_nets2.bin.
4. Build and compile the project. Find the Natural-Selection-Proj.exe file (probably in x64/Debug/), copy it and paste it to the repository folder, replacing the old one.
5. Do the same as in "Run a new simulation with fixed parameters".

References

Evolving AIs - Predator vs Prey, who will win?