Fan Bai

Cooperative perception enables connected autonomous vehicles to extend their sensing range and overcome occlusions by exchanging sensor data. However, its real-world deployment is hindered by asynchronous sensor streams and inaccurate localization of occluded regions. This work presents RAO++ , a real-time cooperative perception system that merges asynchronous sensor data from different vehicles through our novel designs of motion-compensated occupancy flow prediction , on-demand data sharing , with a variety of system optimizations to improve the accuracy and coverage of the perception system. Our comprehensive evaluation, including real-world and emulation experiments under diverse LiDAR configurations, shows that RAO++ outperforms asynchronous-unaware methods by more than 34% in perception coverage and by up to 14% in perception accuracy. Moreover, RAO++ reduces latency by 1.2–3.5× and communication overhead by 45–70% compared with the state-of-the-art asynchronous-aware baseline, while maintaining comparable detection accuracy. Finally, RAO++ demonstrates a practical data overhead of 12.8 KB per frame, enabling deployment under realistic bandwidth constraints.

Accurately predicting 5G bandwidth in vehicular mobility scenarios is essential for optimizing real-time communication (RTC) systems in a range of emerging vehicular applications, such as autonomous driving, in-car video conferencing, vehicular augmented reality (AR), and remote driving in dynamic urban environments. In this paper, we propose Prophet, a causality-aware Transformer model designed to forecast 5G throughput by capturing the complex causal relationships between control-plane 5G network events (e.g., handovers) and environmental factors (e.g., signal strength). Our key contributions include: (1) a wavelet-based encoder that efficiently captures long-term trends through multi-scale signal decomposition; (2) a control-plane causality attention mechanism in the decoder that focuses on localized, short-term throughput variations triggered by network events; and (3) a Granger causality attention mechanism that enhances prediction accuracy by emphasizing historically significant data patterns. Prophet surpasses State-Of-The-Art (SOTA) models like random forest, Long Short-Term Memory (LSTM), and other SOTA transformers in both accuracy and scalability.

Accurately predicting 5G bandwidth in vehicular mobility scenarios is essential for optimizing real-time communication (RTC) systems in a range of emerging vehicular applications, such as autonomous driving, in-car video conferencing, vehicular augmented reality (AR), and remote driving in dynamic urban environments. In this paper, we propose Prophet, a causality-aware Transformer model designed to forecast 5G throughput by capturing the complex causal relationships between control-plane 5G network events (e.g., handovers) and environmental factors (e.g., signal strength). Our key contributions include: (1) a wavelet-based encoder that efficiently captures long-term trends through multi-scale signal decomposition; (2) a control-plane causality attention mechanism in the decoder that focuses on localized, short-term throughput variations triggered by network events; and (3) a Granger causality attention mechanism that enhances prediction accuracy by emphasizing historically significant data patterns. Prophet surpasses State-Of-The-Art (SOTA) models like random forest, Long Short-Term Memory (LSTM), and other SOTA transformers in both accuracy and scalability.

As 5G [3GPP New Radio (NR)] deployments continue to expand, their use as an “opportunistic” bistatic radar sensing system that uses the base station (BS) as the transmitter (TX) and the user equipment (UE) as the receiver (RX) opens the possibility of a new sensing modality that does not need extra hardware, and thus efficiently realize ISAC. In particular, such 5G-based radar sensing can complement and enhance the existing monostatic radar in advanced driver assistance systems (ADAS) and future self-driving cars, e.g., for detecting non-line-of-sight (NLoS) objects. However, 5G-NR signals have been designed for communication purposes and, as per the 3GPP standard definition, they are neither continuous nor periodic in time and frequency. This then makes it challenging to create an efficient standards-compliant bistatic radar system. This paper describes a suite of methods for overcoming these obstacles. We first explore the different reference signals (RSs) defined in the NR standard, and then analyze how they can be best combined for radar purposes. We also present high-resolution parameter-based serial interference cancellation (SIC) to extract the scatterers in the delay-Doppler domain with better-than-Fourier resolution. The impacts of combined precoding and beamforming stipulated in the standard are also discussed in this paper. To demonstrate the validity of our approach, we simulate our results using synthetic channels and validate them with channel measurements.

On-vehicle 3D sensing technologies, such as LiDARs and stereo cameras, enable a novel capability, 3D traffic reconstruction. This produces a volumetric video consisting of a sequence of 3D frames capturing the time evolution of road traffic. 3D traffic reconstruction can help trained investigators reconstruct the scene of an accident. In this paper, we describe the design and implementation of RECAP, a system that continuously and opportunistically produces 3D traffic reconstructions from multiple vehicles. RECAP builds upon prior work on point cloud registration, but adapts it to settings with minimal point cloud overlap (both in the spatial and temporal sense) and develops techniques to minimize error and computation time in multi-way registration. On-road experiments and trace-driven simulations show that RECAP can, within minutes, generate highly accurate reconstructions that have 2× or more lower errors than competing approaches.

5G cellular communication and Multi-access Edge Computing (MEC) have provided a new opportunity to support connected vehicles safety applications complementing Dedicated Short Range Communication (DSRC) and Cellular Vehicle-to-Everything (C-V2X) direct communication. One major benefit of using cellular networks for connected safety is its wide deployment and support in the vehicles today, and it eliminates the need for additional physical radio technologies in vehicles. However, a few major considerations for cellular networks to support real-time safety applications are latency, reliability of communication, cellular costs, and scalability. In this paper, we conduct the first real-world implementation and study of using infrastructure-based camera sensing along with 5G cellular communication and MEC computing setup to support connected safety applications. To achieve this, we design and implement a vision-based multilayer connected safety platform called Raven . We also develop and evaluate three different connected safety applications namely Intersection T-Bone Vehicle Crash Warning, Vehicle-in-blind-spot Warning and Stopped/Disabled Hazard Vehicle Warning applications on top of the Raven platform. These applications are enabled via a roadside camera that is connected via 5G communication link to the MEC server where Raven detects and tracks vehicles in the received camera frames, estimates each vehicle's dynamics such as tracking ID, position, speed, and heading, packages that information along with other parameters required by the Basic Safety Messages (BSM) standard, and shares it with other vehicles that have subscribed to Raven 's service. Our experiments show promising results: the median error in position, velocity, and heading accuracy is $\text{2.5}~m$ , $\text{4.9}~kph$ (or $\text{3}~mph$ ), and $2.1^{\circ }$ , respectively, and the driver is warned in-time during all the test runs of the implemented safety applications.