lc_ngsim's Introduction

LC_NGSIM

lane change trajectories extracted from NGSIM

This dataset comes with the paper below

@inproceedings{dong2017lane,
  title={Lane-change social behavior generator for autonomous driving car by non-parametric regression in Reproducing Kernel Hilbert Space},
  author={Dong, Chiyu and Zhang, Yihuan and Dolan, John M},
  booktitle={Intelligent Robots and Systems (IROS), 2017 IEEE/RSJ International Conference on},
  pages={4489--4494},
  year={2017},
  organization={IEEE}
}

Paper is accepted by IROS 2017

A related Ramp Merging Control paper is published in IV 2017,

here is the accepted version, and an updated version: Ramp merge data and full link to IEEE Xplore will be updated soon. To see the final version:

@inproceedings{dong2017intention,
  title={Intention estimation for ramp merging control in autonomous driving},
  author={Dong, Chiyu and Dolan, John M and Litkouhi, Bakhtiar},
  booktitle={Intelligent Vehicles Symposium (IV), 2017 IEEE},
  pages={1584--1589},
  year={2017},
  organization={IEEE}
}

The original NGSIM program:

@article{
author={Alexiadis,Vassili and Colyar,James and Halkias,John and Hranac,Rob and McHale,Gene},
year={2004},
month={08},
title={The Next Generation Simulation Program},
journal={Institute of Transportation Engineers.ITE Journal},
volume={74},
number={8},
pages={22-26},
}

Source Code.

For source code, click here

Data description

nlc_data_5zones.mat

nlc_data: cell, contains 870 sequences of none-lane-change scenarios.
- veh_s: subjcet vehicle
  - x: Lateral position (m)
  - y: Longitudinal position (m)
  - len: Vehicle length (m)
  - wid: Vehicle width (m)
  - v: Vehicle speed (m/s)
  - a: Vehicle acceleration (m/s^2)
- veh_f: front vehicle in current lane
- veh_r: rear vehicle in current lane
- veh_ft: front vehicle in target lane
- veh_rt: rear vehicle in target lane
- veh_st: overlap vehicle in target lane

lc_data_20s_withpoints.mat

lc_data: Same structure as nlc_data
points: array, n row, 2 column, [start, end]

lc_ngsim's Issues

Is the acceleration obtained by processing the velocity through a second order differential filter?

I use the second-order differential filter to process the velocity and find that the acceleration obtained is very close to the acceleration in the data set.

labels and rkhs application

i have a few questions and have possibly found an error in the script TrajPreProcess.m.

the publication says that the inputs have dimensionality ( 6 (speeds), 6 (longitudinal positions), 6 (lateral positions) )*30 (steps), 6 (car widths) = 541 and the outputs have dimensionality 4 (start and end longitudinal and lateral positions). why does the paper mention in section III.B that the dimensionality of the output is $D=2$ and not $D=4$ as expected (start and end longitudinal and lateral positions imply 4 real numbers)?
the script TrajPreProcess.m script has the code Label(i,1) = trajs.veh_s.y(points(i,1)) - origin_y; Label(i,2) = trajs.veh_s.y(points(i,2)) - origin_y; but why does it not generate the store the labels corresponding to the lateral positions in the matrix Label? this seems to have to do with the question in the point above, that is, the dimensionality of the output should be 4 and not 2, as incorrectly mentioned in the section III.B of the paper.

thanks for your attention!

Recommend Projects