Inspiration
We were inspired by seeing videos of AI powered simulations doing cool things, such as parking in a full parking lot: https://www.youtube.com/watch?v=VMp6pq6_QjI
What it does
Crashes into walls and drives off the edge of the world
How I built it
Trial and error
Challenges I ran into
We had a difficult time getting the reward function correct and tuning the hyperparameters to allow the AI to learn. Although some of has have experience with supervised learning and machine learning for computer vision, we have never tried reinforcement learning before so all of the hyperparameters and details of the training were quite foreign
Accomplishments that I'm proud of
Getting the car to move in the general direction of the goal (sometimes).
It's also really cool to see the AI sometimes take advantage of glitches in the environment and phase through multiple obstacles.
What I learned
Reinforcement learning is difficult and not just a black box.
Log in or sign up for Devpost to join the conversation.