Shifted Power Law

Unveiling Uniform Shifted Power Law in Stochastic Human and Autonomous Driving Behavior

Preparation

1. Create python environment

conda create -n SPL python=3.10
conda activate SPL
pip install -r requirements.txt

2. Download the training and simulation code

# Download the simulator
mkdir shifted_power_law
cd shifted_power_law
git clone https://github.com/CATS-Lab/Shifted_Power_Law.git

3. Download data

You may download the highD, CitySim, and AV datasets for training or simulation.

Training

Illustration of the state for predicting the acceleration of the ego vehicle.

Train the model to predict mean and standard deviation

cd prediction
python train.py --loc_id 4

Simulation

Overview of the agent–based simulation framework.

Run simulations and visualize the results

cd simulation
python main.py --loc_id 4 --data_path ../data/07_tracks.csv --res_root results --vis True --num_sim 1 --dist power_law --init_frame_id 0 --max_frame_id 250

Note: If you do not assign --init_frame_id and --max_frame_id, the simulator will sample the initial frame id randomly and set the last frame id as the maximum frame id.