Abstract: Connected and automated vehicles (CAVs) are a transformative technology expected to improve the safety and efficiency of mobility. As the primary functional components of CAVs, advanced sensing technologies and control algorithms—which gather environmental information, process data, and control vehicle motion—are of paramount importance. Recent years have seen a surge in the development of novel sensing technologies, enabling CAVs to interpret sensory information for obstacle detection, localization, navigation, and interaction within dynamic environments. Furthermore, leveraging computer vision and other sensing methods, in-cabin recognition of human activities, emotions, and mental states is now possible. This Special Issue gathers contributions illustrating the latest advancements in the sensing and control of CAVs.
Introduction: The Challenge of Urban Autonomy
While autonomous vehicles have demonstrated significant capabilities on structured highways, the "urban jungle" presents a far more complex set of challenges. Current Connected and Automated Vehicle (CAV) systems often struggle with sensor latency, scene ambiguity (often termed "casual confusion"), and the simplified physics models typically used in control algorithms. This research collection, curated from a Special Issue of the journal Sensors, explores the "last mile" of vehicle autonomy, focusing on making CAVs safe and efficient in unpredictable urban environments.
Methodology: A Full-Stack Approach
The research presented adopts a comprehensive approach to autonomy, spanning low-level dynamics to high-level coordination.
High-Fidelity Dynamics and Control
To address emergency scenarios, such as high-speed tire failures, researchers developed a 14 Degrees-of-Freedom (DOF) vehicle dynamics model. This model accounts for suspension and tire compression, allowing controllers to maintain stability where traditional 8-DOF models fail. Furthermore, improvements in Model Predictive Control (MPC) were achieved by implementing a backward Euler integration method for predictive modeling, which significantly enhances tracking accuracy without a prohibitive increase in computational overhead.
Sensing and Perception
In the realm of perception, the papers highlight the use of Ultra-Wideband (UWB) technology fused with Dead Reckoning (DR). This combination provides centimeter-level relative positioning accuracy, essential for close-proximity driving in environments where GPS signals are unreliable, such as tunnels. For visual processing, the introduction of Attention Feature Fusion Blocks (AFFB) allows neural networks to prioritize small or distant objects, mimicking human visual focus and improving detection in cluttered scenes.
Human-Centric Integration
Uniquely, the research also explores the human element. One study utilized Electroencephalogram (EEG) data from 35 subjects to correlate cognitive executive control with driving performance. These insights suggest that future CAVs could dynamically adjust their level of autonomy based on the driver's real-time mental state and attention levels.
Experimental Validation and Datasets
The findings are supported by a rigorous mix of industry-standard benchmarks and custom experimental data:
- Computer Vision: Algorithms were trained and validated using the BDD100K and PASCAL VOC datasets, specifically targeting small object detection.
- Traffic Simulation: 24 hours of real-world New York City traffic data was used to simulate and validate urban platooning protocols.
- Physical Testing: Data was gathered from high-speed test circuits (such as the Korea Automobile Testing & Research Institute) to refine vehicle dynamic parameters.
Key Results and Insights
The integration of these advanced methods yielded substantial performance gains across multiple domains:
- Tracking Accuracy: Switching to backward Euler integration in MPC reduced lateral tracking errors by up to 78%.
- Detection Performance: The AFFB approach increased the Mean Average Precision (mAP) for small objects by 3.5%.
- Platoon Efficiency: The Distributed Urban Platooning Protocol (DUPP) reduced average travel time by 20% compared to the current European "ENSEMBLE" standard.
- Safety: The 14-DOF MPC controller successfully prevented rollovers during simulated tire blow-outs at speeds of 101 km/h, a scenario where standard models typically fail.
Conclusion and Future Directions
The research underscores that the future of CAVs lies in the synergy between high-fidelity modeling and intelligent sensor fusion. While challenges remain—specifically regarding the computational load of high-DOF models and the generalization of deep learning models to entirely new urban environments—these studies provide a robust framework for the next generation of autonomous mobility. Future research is directed toward fault-tolerant steering systems and more robust acceleration predictors to further reduce sensor latency and improve safety.
Summary updated