Inspiration
In my country(Nigeria), it's not very surprising to hear about an accident, it's actually sometimes anticipated(in a sense that people already know it's gonna happen). There's a place in Ogun state, a state in Nigeria, where people die literally everyday due to very poor traffic regulation, it's extremely congested. I haven't been there myself, I'd probably die, but a classmate of mine has to pass that route to get to school everyday and he misses school on a lot of days.
What it does
It's supposed to use sensors all over the road network to determine the most efficient way to light up the traffic lights. But since this is a demo, its only going to be a single intersection.
How we built it
I engineered an Autonomous Traffic Control System using Deep Reinforcement Learning (PPO) to optimize 4-way independent intersections, achieving a 43.2% efficiency improvement over traditional fixed-timer systems. By building a custom Gymnasium environment and a real-time Pygame visualizer, I created a "brain" that balances vehicle throughput with pedestrian safety through a weighted reward and penalty system.
Challenges we ran into
-I kept losing the information I had written down since the site didn't save it. So in short, I had -a numpy build error and I didn't know how to fix it until I gave it to Gemini. -I kept on have SSL certificate issues, since I was using MinGW64, I couldn't find a pre-built binary wheel on PyPI. Eventually I just redownloaded python since it just wasn't working and then I messed a lot of my directories. -Internet was slow(I live in Nigeria), so I had to change the command to give it a longer "fuse". I kept trying and trying to break out of using the msys2 python and use the one I downloaded using Winget. Finally time to start training, there were some minor issues though. Broke my terminal. -Made the simulation but it was running too fast to be comprehended. When I adjusted everything to real time the cars started spawning in too fast and when I fixed that, their movement became glitchy(it didn't properly calculate their locations so they stopped at seemingly random positions. All the vehicles would just stop at the intersection. -The reward function wasn't working well i.e it would leave empty roads on green for way too long.
Accomplishments that we're proud of
-I'm proud that I built my first working AI model. -I made my first Github repository. -I completed my first hackathon project.
What we learned
I learned a lot of new commands. -I learned how to use Git and Github. -I learned to properly manage all my files and documents and their directories so I don't run into more problems.
What's next for Traffic Light Regulator
I feel it can be scaled and implemented in other types of intersections and also improved since this is just a demo.
Log in or sign up for Devpost to join the conversation.