Reachability Analysis for Occlusion Reasoning in Realistic Autonomous Driving Scenarios

Abstract: In this Bachelor Thesis, we tackle a crucial issue in autonomous driving technology - the detection and reasoning of occluded areas and potential occluded vehicles in realistic scenarios. Autonomous driving has gained significant momentum in recent years, but ensuring safety remains a major concern. One significant challenge is the presence of objects that obstruct the vehicle’s vision, leading to occlusions that lead to uncertainty for the autonomous system or, in a worse scenario, to unpredicted dangerous situations. To address this problem, we have developed a reachability analysis framework that employs advanced detection methods to ensure safety in autonomous driving scenarios. Our framework focuses on two key aspects: real-time detection of occluded areas and tracking potential phantom vehicles. To evaluate the effectiveness of our approach, we utilized the CARLA simulator, which provides a realistic virtual environment for creating and testing various autonomous driving scenarios. Through experimentation in different situations, we validated the overall performance of our implemented methods. The findings of our project are highly relevant and meaningful to a wider audience. Imagine you’re driving on a busy road, and suddenly, your vision is partially blocked by a large truck or a building. In such situations, our framework can accurately detect the occluded area in real-time, allowing the autonomous driving system to adjust its behavior accordingly. This enhances safety for both autonomous vehicles and human drivers by enabling them to anticipate potential hazards in obscured areas. The impact and implications of our project results extend beyond the realm of autonomous driving. By addressing the issue of occlusions and enhancing safety, our framework contributes to building trust in autonomous technology. This, in turn, promotes its wider adoption and acceptance by society. Moreover, our work aligns with the United Nations’ Sustainable Development Goals, particularly Goal 3 (Good Health and Well-Being) and Goal 9 (Industry, Innovation, and Infrastructure). By ensuring safer roads through advanced detection and reasoning techniques, we contribute to reducing accidents and improving the overall well-being of individuals. In conclusion, our project addresses the critical challenge of occlusions in autonomous driving scenarios. We propose a reachability analysis framework that detects occluded areas and tracks potential phantom vehicles. Our findings have significant implications for safety, trust, and the wider adoption of autonomous technology. By contributing to the advancement of computerassisted diagnostics and aligning with the UN Sustainable Development Goals, our work demonstrates the importance and relevance of researching this field.