Improving behavior profile discovery for vehicles

TL;DR

The paper addresses the challenge of predicting vehicle interactions by extracting driver behavior profiles from undisturbed intersection observations using longitudinal velocity and acceleration. It introduces a framework that combines an Extended Kalman Filter–based similarity measure, an EM-inspired clustering procedure, and KL-divergence–driven split/merge decisions to discover multiple longitudinal behavior profiles for each macro-maneuver without relying on environment maps. Key contributions include a KB-style approach that yields interaction-mode clusters for maneuvers and a mechanism to determine cluster counts via $t_{KL}$. The results on the InD dataset show the method can separate aggressive, normal, and conservative assertiveness profiles and various interaction patterns, enabling more realistic AV prediction and simulation of interactive agents. The work highlights practical challenges, notably the dependence of convergence on $t_{KL}$ and computational cost, pointing to future work in leveraging these clusters to enhance prediction accuracy and AV agent training.

Abstract

Multiple approaches have already been proposed to mimic real driver behaviors in simulation. This article proposes a new one, based solely on the exploration of undisturbed observation of intersections. From them, the behavior profiles for each macro-maneuver will be discovered. Using the macro-maneuvers already identified in previous works, a comparison method between trajectories with different lengths using an Extended Kalman Filter (EKF) is proposed, which combined with an Expectation-Maximization (EM) inspired method, defines the different clusters that represent the behaviors observed. This is also paired with a Kullback-Liebler divergent (KL) criteria to define when the clusters need to be split or merged. Finally, the behaviors for each macro-maneuver are determined by each cluster discovered, without using any map information about the environment and being dynamically consistent with vehicle motion. By observation it becomes clear that the two main factors for driver's behavior are their assertiveness and interaction with other road users.

Figures (12)

• Figure 1: Lon. vel. clusters and centroid heat-maps from \ref{['fig:7b']}
• Figure 2: Example of macro-maneuver (from InD dataset bock2020)
• Figure 3: Two trajectories from different clusters
• Figure 4: Two trajectories from the same cluster
• Figure 5: Clustering architecture proposed