DPGP for Multi-Vehicle Interaction Scenario Extraction

The clustering results on NGSIM and Argoverse are coming soon. <br> This repo provides the python implementation of DPGP algorithm using Gaussian Process to represent multi-vehicle driving scenarios with Dirichlet Process adapting cluster numbers. <br> The python version code is implemented by Mengdi Xu, mengdixu@andrew.cmu.edu @SafeAI lab in CMU. <br> Initial MATLAB code implemented by Yaohui Guo and Vinay Varma Kalidindi. <br>

Paper Reference:

Modeling Multi-Vehicle Interaction Scenarios Using Gaussian Random Field <br> https://arxiv.org/pdf/1906.10307.pdf

Improvement:

(a) fixed several bugs in the MATLAB version of code. <br> (b) The code structure is more clear and can easily be implemented for various applications. <br> Thanks members of SafeAI lab for discussion! <br>

Input:

frames: list with element as object defined in frame.py <br>

Output:

Mixture model as defined in mixtureModel.py <br>

Implement:

Train DPGP: python main_argo.py <br> Visualization: python pattern_vis.py

Required python packages:

argoverse (for lane visualizaiton) <br> numpy == 1.16.4 <br> scipy == 1.3.1 <br> scikit-learn == 0.21.2 <br> pandas == 0.25.0 <br> <Some others: math multiprocessing functools pickle >