Xiaopeng Li

Fang Zhou, Xiaopeng Li, Jiaqi Ma

This paper studies a problem of controlling trajectories of a platoon of vehicles on a highway segment with connected and automated vehicles. This problem is complex because each vehicle trajectory is an infinite-dimensional object and neighboring trajectories have complex interactions (e.g., car-following behavior). A parsimonious shooting heuristic algorithm is proposed to construct vehicle trajectories on a signalized highway segment that comply with boundary conditions for vehicle arrivals, vehicle mechanical limits, traffic lights and vehicle following safety. This algorithm breaks each vehicle trajectory into a few sections and each is analytically solvable. This decomposes the original hard trajectory control problem to a simple constructive heuristic. Then we slightly adapt this shooting heuristic algorithm to efficiently solve a leading vehicle problem on an uninterrupted freeway. To study theoretical properties of the proposed algorithms, the time geography theory is generalized by considering finite accelerations. With this generalized theory, it is found that under mild conditions, these algorithms can always obtain a feasible solution to the original complex trajectory control problem. Further, we discover that the shooting heuristic solution is a generalization of the solution to the classic kinematic wave theory by incorporating finite accelerations. We identify the theoretical bounds to the difference between the shooting heuristic solution and the kinematic wave solution. Numerical experiments are conducted to verify the theoretical results and to draw additional insights into the potential of trajectory control in improving traffic performance. Building upon this foundation, an optimization framework will be presented in a following paper as Part II of this study.

Omar Hussain, Amir Ghiasi, Xiaopeng Li

Connected autonomous vehicles (CAV) technologies are about to be in the market in the near future. This requires transportation facilities ready to operate in a mixed traffic environment where a portion of vehicles are CAVs and the remaining are manual vehicles. Since CAVs are able to run with less spacing and headway compared with manual vehicles or mixed traffic, allocating a number of freeway lanes exclusive to CAVs may improve the overall performance of freeways. In this paper, we propose an analytical managed lane model to evaluate the freeway flow in mixed traffic and to determine the optimal number of lanes to be allocated to CAVs. The proposed model is investigated in two different operation environments: single-lane and managed lane environments. We further define three different CAV technology scenarios: neutral, conservative, and aggressive. In the single-lane problem, the influence of CAV penetration rates on mixed traffic capacity is examined in each scenario. In the managed lanes problem, we propose a method to determine the optimal number of dedicated lanes for CAVs under different settings. A number of numerical examples with different geometries and demand levels are investigated for all three scenarios. A sensitivity analysis on the penetration rates is conducted. The results show that more aggressive CAV technologies need less specific allocated lanes because they can follow the vehicles with less time and space headways.

Pengyue Jia, Yiding Liu, Xiaopeng Li et al.

Worldwide geolocalization aims to locate the precise location at the coordinate level of photos taken anywhere on the Earth. It is very challenging due to 1) the difficulty of capturing subtle location-aware visual semantics, and 2) the heterogeneous geographical distribution of image data. As a result, existing studies have clear limitations when scaled to a worldwide context. They may easily confuse distant images with similar visual contents, or cannot adapt to various locations worldwide with different amounts of relevant data. To resolve these limitations, we propose G3, a novel framework based on Retrieval-Augmented Generation (RAG). In particular, G3 consists of three steps, i.e., Geo-alignment, Geo-diversification, and Geo-verification to optimize both retrieval and generation phases of worldwide geolocalization. During Geo-alignment, our solution jointly learns expressive multi-modal representations for images, GPS and textual descriptions, which allows us to capture location-aware semantics for retrieving nearby images for a given query. During Geo-diversification, we leverage a prompt ensembling method that is robust to inconsistent retrieval performance for different image queries. Finally, we combine both retrieved and generated GPS candidates in Geo-verification for location prediction. Experiments on two well-established datasets IM2GPS3k and YFCC4k verify the superiority of G3 compared to other state-of-the-art methods. Our code and data are available online for reproduction.

Binrui Zeng, Bin Ji, Xiaodong Liu et al.

As Large Language Models (LLMs) demonstrate exceptional performance across various domains, deploying LLMs on edge devices has emerged as a new trend. Quantization techniques, which reduce the size and memory requirements of LLMs, are effective for deploying LLMs on resource-limited edge devices. However, existing one-size-fits-all quantization methods often fail to dynamically adjust the memory requirements of LLMs, limiting their applications to practical edge devices with various computation resources. To tackle this issue, we propose Layer-Specific Adaptive Quantization (LSAQ), a system for adaptive quantization and dynamic deployment of LLMs based on layer importance. Specifically, LSAQ evaluates the importance of LLMs' neural layers by constructing top-k token sets from the inputs and outputs of each layer and calculating their Jaccard similarity. Based on layer importance, our system adaptively adjusts quantization strategies in real time according to the computation resource of edge devices, which applies higher quantization precision to layers with higher importance, and vice versa. {Experimental results show that LSAQ consistently outperforms the selected quantization baselines in terms of perplexity and zero-shot tasks. Additionally, it can devise appropriate quantization schemes for different usage scenarios to facilitate the deployment of LLMs.

Zhixuan Zhang, Xiaopeng Li, Qi Liu

Background subtraction (BGS) is utilized to detect moving objects in a video and is commonly employed at the onset of object tracking and human recognition processes. Nevertheless, existing BGS techniques utilizing deep learning still encounter challenges with various background noises in videos, including variations in lighting, shifts in camera angles, and disturbances like air turbulence or swaying trees. To address this problem, we design a spiking autoencoder network, termed SAEN-BGS, based on noise resilience and time-sequence sensitivity of spiking neural networks (SNNs) to enhance the separation of foreground and background. To eliminate unnecessary background noise and preserve the important foreground elements, we begin by creating the continuous spiking conv-and-dconv block, which serves as the fundamental building block for the decoder in SAEN-BGS. Moreover, in striving for enhanced energy efficiency, we introduce a novel self-distillation spiking supervised learning method grounded in ANN-to-SNN frameworks, resulting in decreased power consumption. In extensive experiments conducted on CDnet-2014 and DAVIS-2016 datasets, our approach demonstrates superior segmentation performance relative to other baseline methods, even when challenged by complex scenarios with dynamic backgrounds.