Event

Trustworthy AI from a Traffic Safety Perspective 
Speakers: Else-Marie Malmek, founder Malmeken AB & Kristina Lillieneke, founder Blackbird Law AB

The pre-study ”Trustworthy AI from a Traffic Safety Perspective”  was initiated by Malmeken AB (Else-Marie Malmek, PM) and Blackbird Law AB (Kristina Lillieneke). The Project Owner was Zenseact AB and the project was supported by a PhD student at VCC, Revere and SAFER. 

The study was aimed at identifying ethical and legal challenges in relation to personal data handling in AI applications and through autonomous driving and also to give suggestions on how we can continue to work with, investigate and solve these issues which are preconditions for us to be able to create Trustworthy AI. We have come to realize that Trustworthy AI is a relatively unexplored area within the automotive industry and the pre-study was initiated since we realized the importance of the issues and had identified several possible legal and ethical risks. 

AVs are part of the solutions to achieve the Vision Zero but with AI solutions there also comes the risks to personal data due to the handling of a huge amount of data gathered by the AVs as well as from its surroundings.

There are enormous benefits to be had from big data analytics, but such analytics also highlights serious privacy problems and there is massive potential for unwanted exposure, monitoring and tracking that can result in anything from embarrassment, discrimination to controlling behaviors. As vehicles are becoming increasingly geared towards being service platforms, behavioral insights can be converted into direct revenue for premium services, infotainment offers, or even partnerships with third parties. But such data collection is not always either legal or ethical and risk hurting brands.

We have identified a number of primary legal and ethical challenges that must be addressed by OEMs and suppliers alike if they want to stay competitive and gain customer trust. Without trust, no business success.  During this seminar we will disclose and discuss some of the important issues to work on. The lecture aims at highlighting these issues and give knowledge which in turn can form the basis for future cooperation and development enabling Trustworthy AI.

Generating Synthetic Scenarios to Test an Al-Enabled Traffic Measurement System
Speakers: Elias Sjöberg, RISE & Lund University & Markus Borg, Lund University 

To make the future of the transport system efficient, safe and sustainable, understanding of the complexities of traffic is vital. Viscando provides detailed traffic data by utilizing Machine Learning and computer vision  algorithms in their infrastructure sensor OTUS3D. The purpose of this study was to test OTUS3D using simulated traffic scenarios. 

Using the MathWorks package RoadRunner, we created a digital model based on a real-life junction located in Lindholmen, Gothenburg. This digital model was imported into the open-source traffic simulator CARLA. Two models were developed to generate parameter values that in turn were used to create different traffic scenarios: one of the models used random sampling while the other was based on the  genetic algorithm NSGA-II. For both models, we investigated the mean error distances, mean speed value errors and  the mean number of misclassifications for simulated trajectories. This provided a proof-of-concept that it is possible to test OTUS3D with  simulated scenarios generated by NSGA-II. 
__________________________________________________________________
 

All SAFER Thursday lunch seminars will be held at the SAFER office and online, via Microsoft Teams, exclusively for SAFER partners. Missing your invitation, please contact Mikael von Redlich.