# Playing a 2D Game Indefinitely using NEAT and Reinforcement Learning

**Authors:** Jerin Paul Selvan, Pravin S. Game  
**arXiv:** [2207.14140](https://arxiv.org/abs/2207.14140)  
**Date:** 2022-07-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2207.14140v1/

## Key points

- Robotics and artificial agents are increasingly common in recent years
- Artificial environments with artificial agents have been created to test new path finding or search space optimization algorithms
- Games like "Flappy Bird" provide an effective way to test these algorithms compared to simulations
- Two algorithms enforced on artificial agents in "Flappy Bird" are NEAT and Reinforcement Learning
- NEAT considers an objective function, crossover, mutation, and augmenting topologies while Reinforcement learning remembers the state, action taken at that state, and reward received for the action taken using a single agent and a Deep Q-learning Network
- Researchers found that changing the initial population size in "Flappy Bird" increased the average score gained by the agent
- The use of NEAT and Reinforcement Learning algorithms has shown promising results in improving agent performance
- Further research could explore other potential algorithms involving multi-dimensional image problems or hidden layers.

## Summary

In recent years, the use of robotics and artificial agents has become increasingly common. To address the challenge of testing the performance of new path finding or search space optimization algorithms, artificial environments with artificial agents have been created. Games such as "Flappy Bird" provide an effective way to test these algorithms by allowing for more efficient determination of algorithm performance compared to simulations. Two algorithms that have been enforced on artificial agents in "Flappy Bird" are NeuroEvolution of Augmenting Topologies (NEAT) and Reinforcement Learning. The NEAT algorithm follows genetic algorithms by considering an objective function, crossover, mutation, and augmenting topologies. Reinforcement learning remembers the state, action taken at that state, and reward received for the action taken using a single agent and a Deep Q-learning Network. Through various iterations by changing the initial population size in "Flappy Bird," researchers found that the average score gained by the agent increased. In conclusion, games like "Flappy Bird" provide an effective way to test path finding or search space optimization algorithms through artificial environments with artificial agents. The use of NEAT and Reinforcement Learning algorithms has shown promising results in improving agent performance; further research could explore the potential of other algorithms such as those involving multi-dimensional image problems or hidden layers.