Machine Learning for Traffic Behavior Analysis in Smart Cities
Team Members: Haejin Song, Dominic Catena
Project Goal
The goal of this project is to create a multi-agent driving simulation benchmark, for facilitating a safer and smarter city environment.
This uses CARLA simulator in order to simulate virtual environments and create digital twins of real-world data.
https://github.com/Dcatna/MARL-Benchmark.git
Week 1
https://docs.google.com/presentation/d/1f8hvH8DlrfTjjbt9Lmyn9MJRMBI33W4QG7BgTtFB0Uk/edit?usp=sharing
Originally, we were on the LLM Smartspace project, so this week was spent learning about the basics of LLMs and DSPy.
Week 2
https://docs.google.com/presentation/d/1n2imZcDlyM54tPsWyb_RU5eez9J0Hve9Z-UexXlSeQ0/edit?usp=sharing
Our focus then shifted to smart cities, which meant we had to familiarize ourselves with Carla and practice installing it on an Orbit node, as we were to install Carla on a fresh Linux machine.
Week 3
https://docs.google.com/presentation/d/1kBu9MbnzAz1UhHuUBWAcpY3UUYn9behWYmEOJaVwzAc/edit?usp=sharing
We took some time to learn about ML/RL concepts and set up Carla on the Linux machine.
Week 4
https://docs.google.com/presentation/d/1UdZb1Ze4KGazArBDfzxaF2YLxJT1vN05b0Kbc6EMbdk/edit?usp=sharing
In order to facilitate the usage of a digital twin, we started setting up VR capabilities to use with Carla, as well as moving equipment to the lab with the Linux machine.
Week 5
https://docs.google.com/presentation/d/1RnvkBB-Y2xWjqLXt5uSasPg-4D36868FGSstfKLfO6k/edit?usp=sharing
Figured out how to access Carla from a different machine to another machine remotely, and made a pedestrian control script.
Week 6/Week 7
Week 8/Week 9/Week 10
https://docs.google.com/presentation/d/1nqpuDp0qs3VF9lKyh8MiSZ6BhWQNiD362zP6yf9rUtE/edit?usp=sharing
https://docs.google.com/presentation/d/1Sg_wYbp9Fa7-YrBQYv9gSuXaD-_rZp0FFWc76_YcjKo/edit?usp=sharing
https://docs.google.com/presentation/d/1JjPmKSONQmfjD1UqSr78W0YUOfYGsWsGFK0q0LeyKKc/edit?usp=sharing
These weeks we worked on setting up the Benchmarking functions, implementing the MAPPO reinforcement learning algorithm, and preparing for the Open House.
