Dataset Source

The challenge dataset is derived from the Open-PAV dataset, maintained by the IEEE ITSS Emerging Transportation Technology Testing (ET3). Open-PAV integrates 14 open-source AV trajectory datasets and provides more than 50 hours of longitudinal trajectories at 10 Hz.

The source datasets cover multiple test sites in the United States and Europe. Additional details on the source data and processing methodology are available in A unified longitudinal trajectory dataset for automated vehicle.

The final challenge dataset can be downloaded here: Download the challenge dataset.

Geographic overview map of source datasets used by the Open-PAV dataset
Figure 1. Overview of the source datasets.

Trajectory Definition

In this challenge, one trajectory segment corresponds to a longitudinal car-following sequence for one following automated vehicle (FAV) and one lead vehicle (LV) in the same lane. The main variable relationships are illustrated in Figure 2.

  • Trajectory-segment index set: I={1,…,i,…,I} \mathcal{I} = \{1, \ldots, i, \ldots, I\}
  • Time index set for segment ii: Ti={ti0,…,tiTi} \mathcal{T}_i = \{t_{i0}, \ldots, t_{iT_i}\} , sampled at Δt=0.1 s \Delta t = 0.1\ \mathrm{s}
  • FAV state at time tt: (xitf,vitf,aitf) (x_{it}^{f}, v_{it}^{f}, a_{it}^{f}) , representing longitudinal position, speed, and acceleration
  • LV state at time tt: (xitl,vitl,aitl) (x_{it}^{l}, v_{it}^{l}, a_{it}^{l})
  • Spatial headway: hit=xitl−xitf h_{it} = x_{it}^{l} - x_{it}^{f} (center-to-center distance)
  • Spatial gap: git=hit−Ll+Lf2 g_{it} = h_{it} - \dfrac{L^l + L^f}{2} (bumper-to-bumper distance)
  • Relative speed: Δvit=vitl−vitf \Delta v_{it} = v_{it}^{l} - v_{it}^{f}

All trajectories are segmented into fixed 30-second intervals.

Trajectory variable diagram 1 Trajectory variable diagram 2 Trajectory variable diagram 3
Figure 2. Definition of trajectory variables.

Variable Definitions

NotationLabelDescriptionFormulationUnit
iiTrajectory_IDTrajectory-segment identifieri∈I i \in \mathcal{I} N/A
ttTime_IndexTime index within trajectory segment iit∈Ti t \in \mathcal{T}_i s
-ID_LVLead vehicle IDIdentifier of the LV paired with the FAV in segment iiN/A
-Type_LVLead vehicle type1 for automated vehicle, 0 for human-driven vehicleN/A
xitlx_{it}^{l}Pos_LVLongitudinal position of the LVxitl, i∈I, t∈Ti x_{it}^{l},\ i \in \mathcal{I},\ t \in \mathcal{T}_i m
vitlv_{it}^{l}Speed_LVSpeed of the LVvitl=xi,t+1l−xitlΔt v_{it}^{l} = \dfrac{x_{i,t+1}^{l} - x_{it}^{l}}{\Delta t} m/s
aitla_{it}^{l}Acc_LVAcceleration of the LVaitl=vi,t+1l−vitlΔt a_{it}^{l} = \dfrac{v_{i,t+1}^{l} - v_{it}^{l}}{\Delta t} m/s2^2
-ID_FAVFollowing automated vehicle IDIdentifier of the FAV in segment iiN/A
xitfx_{it}^{f}Pos_FAVLongitudinal position of the FAVxitf, i∈I, t∈Ti x_{it}^{f},\ i \in \mathcal{I},\ t \in \mathcal{T}_i m
vitfv_{it}^{f}Speed_FAVSpeed of the FAVvitf=xi,t+1f−xitfΔt v_{it}^{f} = \dfrac{x_{i,t+1}^{f} - x_{it}^{f}}{\Delta t} m/s
aitfa_{it}^{f}Acc_FAVAcceleration of the FAVaitf=vi,t+1f−vitfΔt a_{it}^{f} = \dfrac{v_{i,t+1}^{f} - v_{it}^{f}}{\Delta t} m/s2^2
gitg_{it}Spatial_GapBumper-to-bumper distance between LV and FAVgit=xitl−xitf−Ll2−Lf2 g_{it} = x_{it}^{l} - x_{it}^{f} - \dfrac{L^l}{2} - \dfrac{L^f}{2} m
hith_{it}Spatial_HeadwayCenter-to-center distance between LV and FAVhit=xitl−xitf h_{it} = x_{it}^{l} - x_{it}^{f} m
Δvit\Delta v_{it}Speed_DiffRelative speed (LV minus FAV)Δvit=vitl−vitf \Delta v_{it} = v_{it}^{l} - v_{it}^{f} m/s
LlL^l-Length of the LV-m
LfL^f-Length of the FAV-m