RISK ASSESSMENT METHODS IN AUTONOMOUS VEHICLES

Abstract

Autonomous vehicles (AVs) are transforming transportation by allowing cars and trucks to drive themselves. However, urban areas also pose challenges in terms of risk assessment and mitigation in occluded and dynamic driving scenarios. To tackle these challenges on straight as well as curved road segments, the thesis proposes a comprehensive probabilistic risk assessment framework that integrates road geometry computational aspects, particle based forward propagation and uniform distribution models. For straight roads, the framework uniformly samples particles and checks their distance and possible collision with ego vehicle to compute the risks of particles. Also, this framework is developed further in the case of curved roads, with angular velocity calculations being added that would give an idea of how dynamic curvature would impact this assumption of risk. Risk quantification uses variables such as the velocity of objects, and distance from the road up to lateral displacement, allowing for solid estimations on various road types. Results measured in the CarMaker simulation environment validate the scalability and applicability of the method to real-world scenarios. The framework offers insights into how various parameters (particle density, delta time, radius, sigma sensor range, and velocity) impact the accuracy of risk assessment and computation efficiency by iterating through these different parameters. It presents a pathway for joining geometric and probabilistic models of AV risk for safer urban navigation. This has implications for the design of policies and interactions to enhance the reliability and adaptability of autonomous systems.