ASYDE: An Argumentation-based System for classifYing Driving bEhaviors

Vortrag von Giuseppina Monterosso

Veranstaltungsort: APB room 2026
Beginn: 1. Februar 2024 um 11:00
Ende: 1. Februar 2024 um 12:00
Event series: Research Seminar Logic and AI
iCal

Abstract:

Examining the influence of human behavior on road safety is a matter of considerable interest. As studies in this field have demonstrated a correlation between driver behavior and road safety over the past decades, there is a compelling need to develop a software system capable of impartially identifying the typical driving behavior of motorists. The objective of this research is to create one or more algorithms capable of recognizing and classifying a driver's typical driving behavior by integrating various data sources. Since different data sources may present conflicting information due to noise errors or failures, our goal is to resolve such ambiguities by employing artificial intelligence paradigms. We are currently exploring the potential use of a well-known AI framework, namely Dung’s Abstract Argumentation Framework (AAF). Argumentation Frameworks, indeed, provide a valuable tool for analyzing and assessing conflicting pieces of information, enabling us to draw more accurate and dependable conclusions. Our argumentation-based system provides the user with a driving certificate describing their driving behavior. In the certificate, we outline, for each predefined class C of driving behavior (e.g., calm driving, normal driving, aggressive driving), the minimum and maximum percentage of time points where the driving behavior falls in C. The minimum percentage represents the time points where driver behavior has been classified in C without uncertainty, whereas the maximum percentage encodes the time points where some other classes in addition to C have been identified, allowing for multiple interpretations. The experimental evaluation baked up the need for intervals, as for several time points ambiguity in identifying classes occurred, due to different measures collected by different sensors leading to conflicting information. Therefore, assigning a single class is not always possible; instead, multiple possible interpretations need to be considered.

The talk will take place in a hybrid fashion, physically in the APB room 2026, and online through the link:

https://bbb.tu-dresden.de/b/pio-zwt-smp-aus