Dear all,
Hello, I am a college student working with a small group in order to use reinforcement learning to optimize traffic light signals. We have decided to use SUMO for our research, however since we are all unfamiliar with the software we have run into many problems, namely, importing real world traffic sensor data into SUMO to simulate the data in SUMO's gui interface and use the data to train an agent.
The dataset that we are currently working with is from the city of Austin, Texas and provides information on the volume of vehicles as well as the average speed of the vehicles in the last 15 minutes (linked here if it is helpful:
https://data.austintexas.gov/Transportation-and-Mobility/Radar-Traffic-Counts/i626-g7ub/). In doing some research online it appears that the way we are supposed to convert this raw data into usable data for SUMO is to create random trips using the included randomtrips.py and then use calibrators in order to have the amount of vehicles in the simulation align with the volume data from the dataset.
My question is if there is any recent discussion, code example, helpful video/article, or any sort of information/directions that can be linked to or given to me so that I can better understand how this process works. In other words, I am hoping someone is able to better explain how I am able to simulate real traffic sensor data in SUMO or link to me previous discussion/documentation on this topic as I am sure it has been done before. As I stated earlier I, and the rest of my group, am not familiar with SUMO so we are all unsure how to proceed after getting our dataset.
In addition, if anyone has some helpful information/articles/videos or previous discussion on using reinforcement learning or machine learning in SUMO I would greatly appreciate if that could be linked as well. Once we are able to get the data into SUMO, we would have to train an agent to optimize the traffic lights at the given intersection, and we wouldn't have an idea on what that looks like in the sumo environment. I would also love if any other datasets that would be "easier" (more straightforward to import into SUMO) to work with would be shared, as the current dataset is not set in stone.
This project will last until May and I can already predict that we will run into many roadblocks working with SUMO, so I thank you in advance for any help you may be able to provide now and in the future.
Thank you,
Justin
