Scene-Extrapolation: Generating Interactive Traffic Scenarios

TL;DR

This work tackles the challenge of verifying highly automated driving functions by introducing seed-scene extrapolation to generate diverse, interactive traffic futures. A lightweight, multi-agent simulator assigns varied behavior models to actors and evolves scenes in a closed loop, yielding multiple child-scenarios from each seed-scene. Criticality metrics, analyzed per vehicle and across scenarios, are summarized with kernel density estimation to assess a seed-scene's importance for testing HAVs. The approach enables function-agnostic evaluation of test scenarios and reveals that seed-scenes exhibit distinct criticality fingerprints, guiding targeted scenario selection for validation and safety assurance.

Abstract

Verifying highly automated driving functions can be challenging, requiring identifying relevant test scenarios. Scenario-based testing will likely play a significant role in verifying these systems, predominantly occurring within simulation. In our approach, we use traffic scenes as a starting point (seed-scene) to address the individuality of various highly automated driving functions and to avoid the problems associated with a predefined test traffic scenario. Different highly autonomous driving functions, or their distinct iterations, may display different behaviors under the same operating conditions. To make a generalizable statement about a seed-scene, we simulate possible outcomes based on various behavior profiles. We utilize our lightweight simulation environment and populate it with rule-based and machine learning behavior models for individual actors in the scenario. We analyze resulting scenarios using a variety of criticality metrics. The density distributions of the resulting criticality values enable us to make a profound statement about the significance of a particular scene, considering various eventualities.

Figures (6)

• Figure 1: Scene extrapolation: possible futures of a seed-scene are simulated by applying different permutations of behavior models. The resulting child-scenarios are assessed by criticality metrics.
• Figure 2: Block diagram illustrating the simulation process. Trajectory proposals are computed from a seed-scene, considering various behavior models based on the current situation. Each trajectory is followed for a certain timespan before a recalculation is performed, taking into account the changed scene context.
• Figure 3: Example traffic scenes and their corresponding predicted trajectory distribution are shown based on different weightings of the attention value. The weighting of the graph context information used to calculate the trajectory increases from blue to green.
• Figure 4: Comparison of density functions for different sample sizes in regard to the characteristic coverage of the distance metric
• Figure 5: For all plots, the abscissa represents the criticality value and the ordinate represents the density, which is not shown in the plot for visualization purposes. The blue and red colors indicate the distribution of criticality for all child-simulations, the dashed lines marking the measured worst cases in the real situation respectively. If there is no dashed line, the metric could not be calculated for this seed-scene. Additionally, gap time could not be calculated for the red scenario.