“Flying through the obstacles” is a playable game which the model learns to play and trains itself to perform best even in the most difficult levels. The model trains to emerge victorious against human players in any stage. Our model produces the best gaming results that are possible to achieve.
Initially, a game is developed using pygames (add other libraries here) which enables the player to play a game of avoiding to collide with the obstacles encountered on the path. Next, the model uses reinforcement learning to train and learn to play the game in the best achievable way. Once the model completes its learning, it becomes the perfect player and is the toughest competitor to any human player.
- PyCharm
- NeuroEvolution of Augmenting Topologies (NEAT)
- Genetic Algorithm (GA)
- Neural Networks
- Reinforcement Learning
The final result that is achieved from our project is a perfect gaming model, which over the course of its runtime, has made itself learn and adapt, by recognising the patterns, to the ever-changing situations in the game. It basically helps the flying object avoid colliding with the obstacle, hence keep the game going on indefinitely.
Our project has used reinforcement learning as the main tool to train the model on how to perform the best, given the situation and state. Since reinforcement learning is a reward based agent; ergo, it can be used in a number of day to day activities. Self driving car is the most famous and probably the most common application, as it trains the car not to hit any obstacle, instead take necessary actions. It can be of help to the economic institutions, for example predict the prices of the stocks. A traffic signal can also be made efficient using reinforcement learning by tracking the flow of traffic and learning the daily high traffic time, and then turning the signal green for the road with heavy traffic accordingly.
Our project initiates as a fairly simple game, a flying object that tries to avoid hitting any obstacle which would obstruct its path. The aim of the project is to master this game. Therefore, we embedded a reinforcement learning agent in our game, which would experiencand adapt, to the unforeseen states in the game.