CSCS '19: Proceedings of the 3rd ACM Computer Science in Cars Symposium

One of the limiting factors when using deep learning methods in the field of highly automated driving is their lack of robustness. Objects that suddenly appear or disappear from one image to another due to inaccurate predictions as well as occurring perturbations in the input data can have devastating consequences. A possibility to increase model robustness is the use of temporal consistency in video data. Our approach aims for a confidence-based combination of feature maps that are warped from previous time stages into the current one. This enables us to stabilize the network prediction and increase its robustness against perturbations. In order to demonstrate the effectiveness of our approach, we have created a test data set with image perturbations such as image artifacts and adversarial examples in which we significantly outperform the baseline.

We present RobustTP, an end-to-end algorithm for predicting future trajectories of road-agents in dense traffic with noisy sensor input trajectories obtained from RGB cameras (either static or moving) through a tracking algorithm. In this case, we consider noise as the deviation from the ground truth trajectory. The amount of noise depends on the accuracy of the tracking algorithm. Our approach is designed for dense heterogeneous traffic, where the road agents corresponding to a mixture of buses, cars, scooters, bicycles, or pedestrians. RobustTP is an approach that first computes trajectories using a combination of a non-linear motion model and a deep learning-based instance segmentation algorithm. Next, these noisy trajectories are trained using an LSTM-CNN neural network architecture that models the interactions between road-agents in dense and heterogeneous traffic. Our trajectory prediction algorithm outperforms state-of-the-art methods for end-to-end trajectory prediction using sensor inputs. We achieve an improvement of upto 18% in average displacement error and an improvement of up to 35.5% in final displacement error at the end of the prediction window (5 seconds) over the next best method. All experiments were set up on an Nvidia TiTan Xp GPU. Additionally, we release a software framework, TrackNPred. The framework consists of implementations of state-of-the-art tracking and trajectory prediction methods and tools to benchmark and evaluate them on real-world dense traffic datasets.

Evaluating vehicle software and hardware using Hardware in the Loop (HIL) simulation is a very common process in current vehicle manufacturing. However, the more complex the vehicle’s environmental awareness becomes, the more complex the HIL simulation framework has to become. With the introduction of Vehicle-to-Everything (V2X) communication, the environmental awareness of traffic participants expands tremendously. Yet, appropriate tools for evaluating Electronic Control Units (ECUs) with a high level of environmental awareness are lacking. Considering scenarios with more than a handful of vehicles, current HIL simulation frameworks are not capable of simulating these scenarios in real time. Hence, the state-of-the-art testing approach is to provide a non-reactive environment for the validation of V2X ECUs. This paper addresses the question, if and to which extent such a non-reactive approach is sufficient for validating complex V2X based applications.

Automated Vehicle (AV) safety and cybersecurity is an important issue that has to be adequately addressed to ensure that AVs are ready to drive on public roads, and that they are able to safely and efficiently coexist with other motorized and non-motorized traffic participants. So far, there are numerous challenges, such as lack of international standards, software and hardware limitations, absence of methods for integrated safety and security analysis, etc. This paper proposes a novel approach, AVES Framework, for systematic, model-based, integrated AV safety and cybersecurity analysis. AVES Framework adheres to road vehicle development lifecycle and is consistent with international and national standards. It is a flexible method and may be used to analyze any AV regardless of its automation and connectivity level. Furthermore, several relationship matrices and a Safety and Cyber Security Deployment (SCSD) Model, inspired by the Quality Function Deployment method, are used in AVES Framework for relationship analysis and decision making with respect to safety and cybersecurity requirements, measures, and system components.

Intelligent transport systems (ITS) rely on V2X communication for allowing coordination and cooperation of traffic participants and increasing traffic efficiency and safety. Communication between traffic participants needs to be secured, especially with respect to authenticity and integrity. Further, a high level of privacy-preservation needs to be ensured. The current European ITS system relies on the use of pseudonym certificates to achieve trust and security while providing a high level of privacy by ensuring unlinkability of messages in the long term. This paper sets out to investigate whether privacy-preserving attribute-based credentials (ABCs) constitute a viable alternative or complement to the current approach. In particular, this paper focuses on the use case of high-density platooning and investigates the applicability of ABCs in that context.

This paper combines a theoretical model for the risk estimation of a ransomware attack on vehicles with our experiences during an implementation of a real world ransomware as proof of concept. Our gained knowledge on ransomware attacks targeting a real car is transferred into a general model for risk estimation. It provides a generic guideline for the risk estimation and allows for identifying possible weaknesses in a vehicle’s design concerning the threat of automotive ransomware. Through our abstracted approach, this model is applicable on every modern car. An example to prove this model is provided as well.

Automotive electronics is rapidly expanding. An average vehicle contains million lines of software codes, running on 100 of electronic control units (ECUs), in supporting number of safety, driver assistance and infotainment functions. These ECUs are networked using a Controller Area Network (CAN). Security of the CAN bus has not historically been a major concern, however, recent research demonstrate that CAN has many vulnerabilities to cyber attacks. This paper presents a contextualised anomaly detector for monitoring cyber attacks on the CAN bus. Proposed algorithm is based on message sequence modelling, using so called N-grams distributions. It utilises only benign data (one class) for training and threshold estimation. Performance of the algorithm was tested against two different attack scenarios, RPM and gear gauge messages spoofing, using data captured from a real vehicle. Experimental outcomes demonstrate that proposed algorithm is capable of detecting both attacks with %100 accuracy, using far smaller time windows (100ms) which is essential for a practically deployable automotive cyber security solution.