2D Self-Driving Simulator

ShanghaiTech CS181 2024 Spring final project: 2D Self-Driving Simulator

author: Gubin Hu, Jierui Xu, Haojian Wang

If you think the project is inspirational or interesting, please give it a star.

Quick Start

The main idea and methods are detailedly explained in the file "CS181_2D_Self_Driving.pdf".Besides, our project is based on the Car Tracking project of Stanford CS221 Fall 2019-2020, you can also refer to it for gaining a better understanding.

Below are the commands you need to set and control the 2D self-driving simulator.

HMM Agent

To drive manually, run the following command:

python cs181pj-car-hmm/drive.py -l lombard -i none

• You can steer by either using the arrow keys or 'w', 'a', and 'd'. The up key and 'w' accelerate your car forward, the left key and 'a' turn the steering wheel to the left, and the right key and 'd' turn the steering wheel to the right. Note that you cannot reverse the car or turn in place.
• In this phase, no inference has been implemented, so you are unable to see any of the other cars. It is hard to complete the game now.

---

The transfer learned transfer probabilities have already been stored in the ./cs181pj-car-hmm/learned

(Optional) Before driving with inference, transfer probabilities need to be learned by running the following command:

python cs181pj-car-hmm/learn.py -a -d -k 3 -l lombard

To drive with exact inference, run the following command:

python cs181pj-car-hmm/drive.py -a -d -k 3 -l lombard -i exactInference

To drive with a particle filter, run the following command:

python cs181pj-car-hmm/drive.py -a -d -k 3 -l lombard -i particleFilter

• -l specifies the map, with available choices including lombard, small, and small2
• -a specifies to use the agent which follows a fixed path and stops when it detects other cars nearby. After the simulation starts, the agent will remain stationary until it has gathered enough observational data for the HMM to estimate the positions of other vehicles.
• -d specifies to show other cars in the GUI. It cannot be utilized by the agent.
• -k 3 specifies the number of other cars;

Approximate Q learning

To run approximate Q learning, run the following command:

python cs181pj-car-q/qdrive.py -k 3 -l small2 -g -i 50 -t

• -l small2 specifies the map, and small2 is recommended for approximate Q learning;
• -g specifies to show the GUI in the first several iterations;
• -i 50 specifies the number of iterations;

When training is done, it will automatically start trial runs and calculate the rate of success.

Due to the time-consuming nature of training and its inherent randomness, we provide a set of pre-trained weights. To run an approximate Q learning agent with pre-trained weights, remove the -t flag, like this:

python cs181pj-car-q/qdrive.py -k 3 -l small2 -g