The feasibility of infrastructure-vehicle collaborative mapping for intelligent transportation systems

Accurate perception and localisation of dynamic objects is crucial for autonomous vehicles to navigate complex environments and avoid collisions. However, onboard sensing often struggles with complex environments due to the challenges associated with occlusion, hindering progress in achieving advanced levels of vehicle autonomy in these environments. Cooperative driving automation has emerged as an enabling technology for the development of safer vehicles by using shared perception information to enhance the situational awareness of the ego-vehicle. Despite V2I collaboration demonstrating numerous benefits and opportunities, it has not yet been adopted as an industry vertical. This lack of widespread adoption can be attributed to several technical and operational considerations that must be thoroughly understood and addressed before this technology can be successfully implemented on a large scale. To enable the safe deployment of CDA applications, it is essential to investigate how such systems perform under realistic deployment conditions, where factors such as positioning uncertainty, communication delays, and calibration errors are non-zero. This thesis explores the feasibility of V2I map fusion in real-world deployments, amidst the current technological, operational, and regulatory landscape. To address this, the thesis will evaluate enabling technologies, identify deployment challenges, develop a dataset, and quantify the impact of real-world pose alignment errors on map fusion performance. Firstly, a review of the current state-of-the-art for multi-agent and single-agent environmental sensing, perception, and mapping is provided. Then, the practical considerations for V2I CDA deployment in real-world conditions are examined. Following this, a heterogeneous, multimodal dataset captured from the infrastructure perspective is introduced to enable systematic evaluation of CDA technologies. Finally, the robustness of V2I map fusion by analysing the impact of real-world pose alignment errors on V2I CDA is investigated. This thesis shows that deploying V2I systems involves substantial challenges spanning infrastructure sensor selection, edge computing optimisation, policy and privacy issues, and economic viability. Experiments conducted demonstrate that pose alignment errors from communication delay, calibration errors, and positioning uncertainty can significantly degrade collaborative perception reliability. Establishing well-defined operational design domains and error thresholds is paramount to ensuring the safety and reliability of V2I systems as they scale up. To utilise the technology's full potential, a gradual rollout strategy in small controlled use cases is necessary, enabling stakeholders to address technical and regulatory challenges in manageable settings.

Funder

Taighde ´Eireann — Research Ireland
European Commission
Valeo Vision Systems

Publisher

University of Galway

Rights

CC BY-NC-ND

cbn

Collections

University of Galway Theses (PhD Theses)