Open-vocabulary object detection (OVD) is crucial for handling dynamic real-world driving scenarios. Inspired by YOLO-World, we propose OpenVocab-Auto, an open-vocabulary object detection framework with driving-aware multi-scale feature fusion for autonomous driving scenarios. Our system introduces three key innovations: (1) a context-adaptive prompt engine that significantly reduces computational overhead compared to global prompt strategies, (2) hierarchical vision-language alignment for improved small object detection, and (3) real-time optimization achieving 27 FPS on NVIDIA Jetson AGX Orin through TensorRT acceleration. On RTX 3080 (FP16 full model), the framework achieves 0.923 F1 for parking space detection and 0.847 F1 for zero-shot obstacle recognition. On Jetson Orin (TensorRT INT8 model), the corresponding scores are 0.811 and 0.333, respectively, under the same evaluation protocol.

The high mobility characteristics of connected vehicles present noteworthy difficulties in the domain of federated learning. Based on our understanding, current federated learning strategies do not tackle the challenge of continuously training multiple models for vehicles in constant motion, which are subject to variable network conditions and changing environments. In response to this challenge, we have created and implemented iFLOW, a versatile and intelligent multi-model federated learning infrastructure specifically designed for highly mobile-connected vehicles. iFLOW addresses these challenges by integrating four key aspects: (1) a strategically devised model allocation algorithm that dynamically selects vehicle computing units for distinct model training tasks, optimizing for both resource efficiency and performance; (2) a dynamic client vehicle joining mechanism that ensures smooth participation of vehicles, even in the face of signal loss or weak connectivity, mitigating disruptions in the training process; (3) integration of a large language model (Llama3.3 70B) as an intelligent arbiter for decision-making within the framework, enhancing adaptability and robustness; and (4) real-world deployment and testing on distributed vehicular devices to validate the approach. The experimental evaluation demonstrates that iFLOW allows multiple models to train asynchronously and outperform centralized training. These results affirm the effectiveness of iFLOW in practical, real-world scenarios involving highly mobile vehicular networks.

Connected vehicles (CVs) face significant challenges in continuous big data transmission, resulting in high transmission bandwidth costs and impacting real-time decision-making. To address this, we propose two dynamic, driving-aware compression mechanisms based on reinforcement learning and temporal compressive sensing to intelligently compress video data. These mechanisms adapt to driving conditions, reducing bandwidth while preserving sufficient information for accurate applications such as object detection and ensuring high-quality reconstruction when needed. We also implement a Vehicle-EdgeServer-Cloud (VEC) closed-loop framework that integrates these mechanisms. Specifically, a lightweight vehicle model performs real-time detection on compressed data (measurements), while the EdgeServer receives measurements and reconstructs scenes if needed. The measurements, reconstructed video, and analysis results are then sent to the cloud for vehicle model updates. Unlike conventional methods, our framework seamlessly adapts across vehicles, Edge-Servers, and the cloud, supporting efficient data transmission and dynamic model updates. Extensive evaluations were conducted on our designed roadside unit platform and robotic vehicle, both equipped with industry-grade sensors and computing units. The results demonstrate an 18x reduction in bandwidth at 320KB/s while maintaining high detection accuracy and reconstruction quality compared to non-adaptive measurements, highlighting the framework's promising real-world applications for CVs.

In this paper, we demonstrate an Autoware-based autonomous driving system featuring depth sensing within a Virtual Machine (VM). Utilizing off-the-shelf depth cameras and ROS2 for data exchange, we assess the viability of replacing traditional LiDAR systems for environmental perception. Calibration tests reveal a depth accuracy within ± 5 mm. Additionally, we explore the system's portability by deploying it on a Jetson Nano-based miniature vehicle, uncovering computational constraints. The study outlines areas for future optimization and real-world applicability

This paper focuses on addressing computational constraints and energy limitations prevalent in edge-based applications through an innovative approach, dynamic in-situ control for edge-based applications (DICE). DICE capitalizes on the burgeoning trend in vehicle sensor technologies, such as camera, Radar, and LiDAR, which are becoming increasingly powerful and capable of performing pre-processing computations. DICE introduces a concept of "downstream offloading", which distinguishes it from traditional offloading approaches that typically offload computational tasks from edge devices to more powerful Edge Servers. In contrast, DICE offloads part of the computational tasks from the Edge Server to the sensor itself, thereby optimizing data processing at the source and reducing the volume of data transmission required. This approach not only addresses the latency bottleneck frequently encountered in energy-intensive neural networks but also enhances the efficiency of data processing by selectively filtering out non-critical frames based on event-triggering mechanisms. DICE leverages the unique strengths of portable devices such as smartwatches and smartphones, even with their inherent computational and power limitations. The framework consists of an adaptive control layer for dynamic task allocation and an application layer designed to deploy quantized models on System on Chips (SoCs) like TinyML, thereby improving the efficiency of AI-driven applications while conservatively utilizing energy. This system proposes a sustainable, energy-efficient pathway for future edge-based applications.

In this paper, we propose an Android-based distributed computing framework for accelerating DNN inference on Android edge devices. We experimentally demonstrate that the proposed distributed framework can reduce CPU utilization by 24 % (making the the CPU utilization close to that of idle status), reduce power consumption by 59.8 % to 71.8 %, without leading to high-bandwidth througput. The proposed framework can be applied to various Android devices to enable cooperation among edge devices in a distributed computing manner, accelerate DNN inference, and enrich the functionality of Android devices to enhance user experience.

The recent proliferation of computing technologies (e.g., sensors, computer vision, machine learning, and hardware acceleration) and the broad deployment of communication mechanisms (e.g., dedicated short-range communication, cellular vehicle-to-everything, 5G) have pushed the horizon of autonomous driving, which automates the decision and control of vehicles by leveraging the perception results based on multiple sensors. The key to the success of these autonomous systems is making a reliable decision in real-time fashion. However, accidents and fatalities caused by early deployed autonomous vehicles arise from time to time. The real traffic environment is too complicated for current autonomous driving computing systems to understand and handle. In this article, we present state-of-the-art computing systems for autonomous driving, including seven performance metrics and nine key technologies, followed by 12 challenges to realize autonomous driving. We hope this article will gain attention from both the computing and automotive communities and inspire more research in this direction.

The proliferation of edge computing technologies has boosted the development of new applications for a plethora of edge devices. However, many applications face privacy issues and bandwidth limitations. To solve these limitations, we propose a collaborative learning framework on the edges, named CLONE, which is steered by the real-world data sets collected from a large electric vehicle (EV) company and a grocery store of a shopping mall, respectively. We categorize two application scenarios for CLONE, i.e., CLONE in the training stage (CLONE_training) and CLONE in the inference stage (CLONE_inference). As to CLONE_training, we choose the failure prediction of EV battery and associated components as the first use case. While as for CLONE_inference, customer tracking in a grocery store is selected as another case study. In this work, the goal of the CLONE is to support real-time training and inference for connected vehicles and marketing intelligence services. Our experimental results on the EV data show that CLONE is able to reduce model training time without sacrificing algorithm performance. Furthermore, the experimental results on the video data from the grocery store reveal that CLONE is a useful approach to solve the multitarget multicamera tracking problem in a collaborative fashion.

Accurate prediction of supercapacitor's performance under different operating condition is important for its reliable and optimized application. In order to predict the characteristic parameters of supercapacitor under various working condition with small number of experiments, the paper presents a prediction method with less dependency on experimental data. Firstly, a set of experiments have been carried out to get mission profile data. Then, a comparative study of parameter estimation using Support vector machine (SVM) and traditional multiple linear regression (MLP) has been carried out. It's concluded that SVM can achieve same fitting and prediction ability with small sized data. Furthermore, the available energy of supercapacitor energy storage system has been derived based on the prediction results, and the effectiveness of the proposed method is verified by experiments.