AI Path Finding Optimizer

Inspiration

While exploring existing reinforcement path finding algorithms we came across the fact that both Value (Deep Q learning Algorithm) and Policy (Markov Chain Algorithm) driven algorithms have their own pros and cons related to training time, model efficiency and accuracy. These algorithms were disastrous for some gaming applications.

To overcome this we built a path finding algorithm and simulated it using a snake game by combining both Policy and Value driven approach of path finding highlighting pros of each algorithm by reducing the training time and enhancing the model for accuracy and efficiency.

What it does

The model aims to show a comparative analysis between the "Actor Critic Algorithm" and all other existing path finding algorithms, by simulating the behavior of all these algorithms in different environments within the classic snake game.

The model aims to predict better path finding algorithms on the basis of the moves taken to complete the entire game.

How we built it

We built the "Snake Classics" game facilitating Python3, PyTorch, Keras, Tensorflow, PyGame, OpenCV and then integrated the game with "The Actor Critic Algorithm" and all other existing path finding algorithms implementing them.

Challenges we ran into

Some major challenges were integrating 2 completely different learning approaches into one algorithm and training the model. i.e. Combining Value and Policy driven Reinforcement Learning.

However, we achieved better by the developed algorithm.

Accomplishments that we're proud of

We , finally now have a robust algorithm i.e. "The Actor Critic Algorithm" that requires less training data and training time, and operates in infinite space, achieving better accuracy, efficiency and performance in comparison to all the existing path finding algorithms.

What we learned

We understood the drawbacks of existing path finding systems that are limited to operate within finite spaces and also learnt to develop path finding algorithms that are capable of operating in finite as well as infinite spaces i.e Operating when there is a tremendous amount of data.

What's next for The AI Path Finding Optimizer

Further, we plan to widen the scope of this algorithm by combining it with Computer Vision to improve path finding and localization mechanisms in robots and autonomous cars.