Traffic safety is the fundamental criterion for vehicular environments and many artificial intelligence-based systems like self-driving cars. There are places, e.g., intersections and shared spaces, in the urban environment with high risks where vehicles and vulnerable road users (VRUs) such as pedestrians and cyclists directly interact with each other. By advancing starte-of-the-art artificial intelligence methodologies, this project VeVuSafety aims to build a privacy-aware deep learning framework to learn road users’ behaviour in various mixed traffic situations for the safety between vehicles and VRUs. VeVuSafety proposes a 3D environment model based on 3D point cloud for privacy protection — private information like license plates and face is anonymized. Then, within this environment model, an end-to-end deep learning framework using camera data will be built for multimodal trajectory prediction, anomaly detection, and potential risk classification based on deep generative models such as Variational Auto-Encoder. Additionally, an active privacy mechanism will also be adopted by application of the differential privacy mechanism to help the deep learning models prevent model-inversion attack. Moreover, the framework’s generalizability will be investigated by exploring the Normalizing Flows approach for domain adaption. The framework’s performance will be validated at different intersections and shared spaces using real-world traffic data. Besides road user safety and privacy, VeVuSafety can help traffic engineers and city planners to better estimate the design of traffic facilities in order to achieve a road-user-friendly urban traffic environment. Furthermore, the success of VeVuSafety will enhance the fellow’s scientific knowledge and project management skills to become an artificial intelligence expert for traffic safety and Intelligent Transportation Systems.

HYMEDNA prepares commercialization of very sensitive point-of-care biosensors for early-stage cancer detection based on the electrical detection of hypermethylated DNA (hmDNA) inside nanogaps. Only recently the awareness has risen that local hypermethylation of DNA provides a generic marker for a wide range of cancers. A robust, simple and cheap method for detecting hmDNA at low concentrations in blood, urine or faeces would be a major step forward in the early-stage detection of cancer. Existing hmDNA detection relies on fluorescent read-out, which requires dedicated laboratory handling. Our technology is based on electrodes separated by a tunable nanogap. hmDNA is trapped and highly concentrated in between the electrodes using methyl binding domain (MBD) proteins. MBD binds specifically to methylated CpG sequences and thus provides the direct recognition of the targeted methylated moieties, contrasting existing DNA detection methods which commonly employ DNA (or PNA) oligos and rely on sequence specificity. After target binding, the conductivity of the trapped hmDNA is enhanced, for which we provide alternative routes. The detection step is formed by a simple measurement of the electrical conduction between the electrodes. As our detection scheme relies on completely turning on the conduction instead of only modulating it (as in other electrical detection schemes), an exceptionally high sensitivity is expected. As the most competitive advantages we identify (1) high selectivity (specific chemistry) and sensitivity (“on-off” effect), (2) simple, scalable device concept, (3) robustness against environment, (4) small size (allowing for implementation in a bioassay device for multiple target molecules), and (5) low cost price.

Effective tool support for the joint analysis of safety and security risks is long overdue. Risk management is an important activity to ensure the reliable functioning of technology, such as power plants and self-driving cars. Risks include both safety (= accidental failures) and security aspects (= malicious attacks). Historically, safety and security risks have been analyzed in isolation, despite often conflicting with each other. Effective decision-making requires considering safety and security risks in combination, as measures that increase safety may decrease security and vice versa. My Consolidator Grant CAESAR has laid out the groundwork for a safety-security co-analysis framework: (1) A graphical risk model, mapping how vulnerabilities and failures propagate and cause system-level disruptions; (2) efficient algorithms to compute risk metrics, indicating how well a system performs in terms of safety-security. (3) algorithms that quantify the uncertainty of the analysis algorithms. In RUBICON, I will develop a PoC software tool that supports methods from CAESAR, advancing from TRL1 to TRL3. Key challenges to be tackled include: • Scaling up analysis methods to handle industry-size problems, by tailoring algorithms to work with specific subclasses that appear in practice. • Improve the interpretability of calculated outcomes. We will develop diagnostic feedback methods based on counter example analysis and importance factors. • Multi-objective optimization techniques. When dealing with multiple, interdependent parameters, conflicting requirements often arise, due to resource constraints and varying priorities. RUBICON will develop optimal strategies to effectively balance such conflicts, exploiting and advancing Pareto-analysis. The proof-of-concept tool will be tested and validated via lab and pilot studies across different industrial domains. A serious market analysis will lay out an actionable strategy to commercialize the PoC tool post-project.