TrajGAIL

Introduction

Generative model for urban vehicle trajectories based on Deep Learning This repository include implementations of :

• Markov Mobility Chain Model for next location prediction (Gambs et al. 2012)
• RNN based trajectory generator (Choi et al. 2018)
• MaxEnt inverse reinforcement learning (Ziebart et al. 2008)
• TrajGAIL based on Generative Adversarial Imitation Learning (Ho et al. 2016, Choi et al. 2020)
• ShortestPath World (MDP for routing imitations)

Data availability

Due to the public availability issue of taxi data of Gangnam District, it is not possible to upload the taxi data.

The available data is a virtual vehicle trajectory data generated by AIMSUN shortest path routing engine.

Below figure shows the network configuration.

Requirements

python>3.7

required python packages in requirement.txt

pip install -r requirement.txt

How to Run

To run Behavior Cloning MMC Test

python scripts/behavior_clone/run_bc_rnn.py

To run Behavior Cloning RNN Test

python scripts/behavior_clone/run_bc_rnn.py

To run MaxEnt IRL

python scripts/irl/demo_shortestpath.py

To run TrajGAIL

python scripts/gail/run_gail.py

Citations

If you use this code for your research, please cite our paper.

@article{choi2020trajgail,
  title={TrajGAIL: Generating Urban Trajectories using Generative Adversarial Imitation Learning},
  author={Choi, Seongjin and Kim, Jiwon and Yeo, Hwasoo},
  journal={arXiv preprint arXiv:2007.14189},
  year={2020}
}