se-ke/ai-minesweepers

Visual simulation that combines a feedforward neural network with a genetic algorithm to train agents to efficiently pick up blue mines and to avoid bad asteroids.

Hi, I'm Sean👋

I am a software engineer based out of Boston, and I'm currently working as a software development engineer at Amazon Robotics. Before joining Amazon Robotics, I studied computer science and information ethics at Northeastern University, which I graduated from in May 2023. In addition to my experience at Amazon Robotics, I also worked as a backend engineer at a small social media startup called ListHall. I've also worked as a graphics development co-op at Ansys and as a technology research analyst at Accenture Labs. Outside of computing, I enjoy running, rock climbing, and learning to play the guitar.

How to reach me:

• 📫 Email: [email protected]
• 🌐 Personal Website: se-ke.github.io
• 📋 LinkedIn: www.linkedin.com/in/seankellycs
• 😄 Pronouns: He/Him

ai-minesweepers's Issues

Rework getClosestMine to check across the edge of the screen

Currently, getClosestMine works by checking the distance between each agent and each mine as if they were on a flat surface, a rectangle. However, the edges of the rectangle are in fact linked together to form a torus. So, if an agent is on the left side of the screen, it will not determine that a mine on the right side of the screen is closest to it because it does not account for the edges being connected. This could be fixed by testing multiple location of the mine (ex. (x,y), (x-width, y), (x, y-height), (x-width, y-height), (x+width, y), (x, y+height), (x+width,y+height)), but this is rather burdensome.

Button for neural network visualization

Implement a button that when pressed, toggles the neural network visualization section of the draw function in the AgentMS class.