Traffic scenarios and situations that are especially critical and/or occur only very rarely in the real world (i.e. anomalies) are of particular interest for the development and verification of autonomous vehicles. An essential part towards so called scenario-based testing is to extract such scenarios automatically from data sets as a first step and generate them synthetically as a next step, using deep learning methods on graph-structured data. One direction towards finding edge case scenarios in large-scale datasets is to apply concepts of anomaly detection. Specifically, as traffic scenarios are modeled as heterogeneous, spatio-temporal graphs in our research, the problem boils down to exploring and applying graph anomaly detection to especially large and complex scenario graphs.

Literature

• X. Luo et al., “Deep graph level anomaly detection with contrastive learning,” Sci Rep, vol. 12, no. 1, Art. no. 1, Nov. 2022, doi: 10.1038/s41598-022-22086-3.
• X. Wang, B. Jin, Y. Du, P. Cui, and Y. Yang, “One-Class Graph Neural Networks for Anomaly Detection in Attributed Networks,” Neural Comput & Applic, vol. 33, no. 18, pp. 12073–12085, Sep. 2021, doi: 10.1007/s00521-021-05924-9.
• C. Qiu, M. Kloft, S. Mandt, and M. Rudolph, “Raising the Bar in Graph-level Anomaly Detection.” arXiv, May 27, 2022. doi: 10.48550/arXiv.2205.13845.
• E. Meyer, M. Brenner, B. Zhang, M. Schickert, B. Musani, and M. Althoff, “Geometric Deep Learning for Autonomous Driving: Unlocking the Power of Graph Neural Networks With CommonRoad-Geometric.” arXiv, Apr. 24, 2023.

Structure

• Literature research and exploration of the current state-of-the-art in …
  •   … graph representation learning using Graph Neural Networks (GNN)
  •   … graph-anomaly detection, specifically for dynamic and heterogeneous graphs
• Evaluation of existing methods and application to complex traffic scenario graphs
• Improvement and extension of existing methods to the specific domain of scenario-based testing in AD
• Benchmark, quantitative and qualitative evaluation of different approaches
• Composition of a scenario catalog / dataset of semantically “abnormal” traffic scenarios

Preferred skills

• Solid foundations in the field machine learning, specifically deep learning
• Preferably basic knowledge of Graph Neural Networks
• Hands-on experience with Python and frameworks like PyTorch and / or TensorFlow
• Willingness to acquire new technical knowledge, read and understand scientific papers and work independently
• Fluent in German and / or English

Required documents

• Brief cover letter (3-4 sentences)
• Brief CV (max. 2 pages)
• Your current grades (Notenauszug)