Inspiration
Our current traffic light systems are outdated. We wanted to create a new system capable of smart, dynamic traffic control. One of the best ways to decrease road traffic is removing cars from the situation, so in our initial plan we wanted to encourage walking and the use of public transit.
What it does
Our system uses a reinforcement learning agent trained to minimize time spent at lights for high occupancy vehicles, like busses, and pedestrians. Objects are observed by a series of colliders stored inside of the intersection and decisions are evaluated based off of weight. Emergency vehicles, pedestrians, and busses are weighted highest.
How we built it
This system was built in C# in Unity, using the ML agents machine learning package.
Challenges we ran into
Creating the simulation:
- Creating a full simulation capable of observing intersection activity and preventing collisions proved to be especially difficult as the number of vehicles and pedestrians increased. We found a good balance by varying object speed and setting tolerances for entity movement through trial and error.
Log in or sign up for Devpost to join the conversation.