Clustering of the Scenario Space for the Assessment of Automated Driving

Abstract

Assessment and testing are among the biggest challenges for the release of automated driving. Up to this date, the exact procedure to achieve homologation is not settled. Current research focuses on scenario-based approaches that represent driving scenarios as test cases within a scenario space. This avoids redundancies in testing, enables the inclusion of virtual testing into the process, and makes a statement about test coverage possible. However, it is unclear how to define such a scenario space and the coverage criterion.This work presents a novel approach to the definition of the scenario space. Spatiotemporal filtering on naturalistic highway driving data provides a large amount of driving scenarios as a foundation. A custom distance measure between scenarios enables hierarchical agglomerative clustering, categorizing the scenarios into subspaces. The members of a resulting cluster found through this approach reveal a common structure that is visually observable. We discuss a data-driven solution to define the necessary test coverage for the assessment of automated driving. Finally, the contribution of the findings to achieve homologation is elaborated.

UnScenE: Toward Unsupervised Scenario Extraction for Automated Driving Systems from Urban Naturalistic Road Traffic Data

“…This is in accordance with literature, in which HAC is described as suitable for use cases with many clusters and many samples [21]. Furthermore, this is in line with other publications in the context of real-data based scenario extraction, both based on similarly structured data sources [33] as well as for other types of data sources [34].…”

“…In literature, three different testing criteria categories, namely external, internal and relative indices, are defined to be able to estimate the quality of clustering results [20]. In accordance with present work in the field of unsupervised scenario extraction (e.g., [33], [34], and [38]), one branch of our evaluation approach can be assigned to the external indices category, where external information is used as standard to validate the clustering results. For this purpose, we use the map information available within the datasets in the form of images of the corresponding traffic spaces.…”

“…Elpsas et al [17] propose a method, where rule-based detections are used to train fully convolutional networks for extracting lane change and cut-in scenarios. Kerber et al [33] define a custom scenario distance measure, considering the occupancy and relative distances of vehicles in eight slots surrounding the ego-vehicle, for subsequent HAC. The authors mention that the method is not suited for urban scenario extraction without modifications.…”

Novelty Detection and Analysis of Traffic Scenario Infrastructures in the Latent Space of a Vision Transformer-Based Triplet Autoencoder

“…More similar to this work are [3] (LSTM+CNN), [4] (SeqDSPN), [5] (RC-GAN), where the latent representations of deep neural networks are used to cluster scenarios. Also a lot of non-machine learning approaches were used to determine representations or similarities of scenarios, such as in [3] (DTW), [6] (DTW+PCA), [12] (Dynamic-Length-Segmentation), [13] (custom similarity measure). In this work, an expert-knowledge aided latent representation is introduced, and hence differs from the aforementioned works.…”

“…In the works [10], [11] only little static information are used. Whereas in the works [3], [4], [5], [6] and [13] only spatial and dynamic information is used. This work focuses on the static environment of a traffic scenario.…”