7,276 publications found
Sort by
A Demand Truncation and Migration Poisson Model for Real Demand Inference in Free-Floating Bike-Sharing System

The free-floating bike-sharing system enables users to pick up bikes from everywhere. However, the spatial-temporal imbalance of bikes becomes severer in such systems. Consequently, bikes are sometimes not available at desired location and time, and the pick-up demand of users is truncated. On demand truncation, the bike user would either give up using the bike or migrate to the nearby to pick up the bike. Thus, the demand of bikes observed from smart locks does not equate with the real demand. However, capturing the real demand is crucial for system planning and operation. Therefore, a statistical model, namely the demand truncation and migration Poisson (DTMP) model, is proposed to analyze the real demand. It is able to jointly model the real demand of FFBS and the migration behavior of users. In particular, we analyzed the demand truncation and migration processes to establish the relationship between the expected value of observed demand and that of real demand. Then, the real demand of using bikes was assumed to be related to the influential factors and the demand function was formulated. Subsequently, the likelihood for each observation was established. The performance of the proposed model was evaluated through field-testing data. The results revealed that the proposed DTMP model is superior to the baseline Poisson model from both the fitness and accuracy perspectives. Finally, the spatial-temporal distribution of the real demand and unmet demand of bikes in typical hours are presented. The results can guide the efficient management of the bike-sharing system.

Interactive Multi-Scale Fusion of 2D and 3D Features for Multi-Object Vehicle Tracking

Multiple Object Tracking (MOT) is a significant task in autonomous driving. Nonetheless, relying on one single sensor is not robust enough, because one modality tends to fail in some challenging situations. Texture information from RGB cameras and 3D structure information from Light Detection and Ranging (LiDAR) have respective advantages under different circumstances. Therefore, feature fusion from multiple modalities contributes to the learning of discriminative features. However, it is nontrivial to achieve effective feature fusion due to the completely distinct information modality. Previous fusion methods usually fuse the top-level features after the backbones extract the features from different modalities. The feature fusion happens solely once, which limits the information interaction between different modalities. In this paper, we propose multi-scale interactive query and fusion between pixel-wise and point-wise features to obtain more discriminative features. In addition, an attention mechanism is utilized to conduct soft feature fusion between multiple pixels and points to avoid inaccurate match problems of previous single pixel-point fusion methods. We introduce PointNet <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$+$</tex-math> </inline-formula> <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$+$</tex-math> </inline-formula> to obtain multi-scale deep representations of point clouds and make it adaptive to our proposed interactive feature fusion between multi-scale features of images and point clouds. Through the interaction module, each modality can integrate more complementary information from the other modality. Besides, we explore the effectiveness of pre-training on each single modality and fine-tuning on the fusion-based model. Our method can achieve 90.32% MOTA and 72.44% HOTA on the KITTI benchmark and outperform other approaches without using multi-scale soft feature fusion.

Revealing Senior Mobility Patterns and Activities in Urban Transit Systems

Urban mobility plays a crucial role in maintaining seniors’ well-being and quality of life, with urban transit systems serving as their primary mode of travel. This study focuses on revealing seniors’ travel patterns and activities and predicting their demands in transit systems. We propose a spatio-temporal travel pattern model that extends the structural topic model from text mining, to detect senior mobility patterns from trip sequences at the rail station level and discover their activities hidden in passively collected data from a probabilistic generative procedure. The proposed model can incorporate a wide range of spatio-temporal covariates that help estimate the relative importance of travel patterns across stations and over time. Numerical experiments are conducted using transit smart card data collected in Nanjing, China. The results demonstrate that our proposed methodology successfully transforms massive mobility records into informative travel patterns, each of which characterizes a distinct distribution of trip attributes. We combine land use characteristics of station areas with detected patterns to explore seniors’ behavioral strategies and activities at multiple spatial and temporal scales. Moreover, by applying prior detected patterns to the station-level ridership prediction, we significantly improve the predictive performance of direct ridership models. This study contributes to detecting seniors’ travel patterns, discovering latent activities, predicting station-level travel demands, and creating senior-friendly mobility systems.

A Hierarchical Vehicle Behavior Prediction Framework With Traffic Signals and Interactive Agents

Vehicle behavior prediction in complex urban scenarios with traffic signals and interactive agents is an important yet complicated task for autonomous vehicles (AVs). In this work, a hierarchical vehicle behavior prediction framework is proposed to incorporate the traffic signal information and model the interaction between vehicles. The framework predicts vehicle behaviors in two stages, discrete intention prediction and continuous trajectory prediction. In the discrete intention prediction stage, Bayesian network is adopted to provide a high-level behavior prediction of the principle other vehicle. The discrete prediction results are forwarded to the second stage, where a continuous trajectory is predicted with maximum entropy inverse reinforcement learning and potential game. The framework is designed to be able to capture the difference among human drivers with parameterized driver characteristics. The proposed predictor is validated in two scenarios: the yellow light running scenario and the right-turn scenario. The trajectory prediction average displacement error of the yellow light running scenario is 0.695m for a 3-second prediction interval, and the prediction accuracy of the right-turn vehicle in the right-turn scenario is 0.51m for a 2-second prediction interval.

Detecting State of Charge False Reporting Attacks via Reinforcement Learning Approach

The increased push for green transportation has been apparent to address the alarming increase in atmospheric <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${CO}_2$</tex-math> </inline-formula> levels, especially in the last five years. The success and popularity of Electric Vehicles (EVs) have led many carmakers to shift to developing clean cars in the next decade. Moreover, many countries around the globe have set aggressive EV target adoption numbers, with some even aiming to ban gasoline cars by 2050. Unlike their gasoline-based counterparts, EVs comprise many sensors, communication channels, and decision-making components vulnerable to cyberattacks. Hence, the unprecedented demand for EVs requires developing robust defenses against these increasingly sophisticated attacks. In particular, recently proposed cyberattacks demonstrate how malicious owners may mislead EV charging networks by sending false data to unlawfully receive higher charging priorities, congest charging schedules, and steal power. This paper proposes a learning-based detection model that can identify deceptive electric vehicles. The model is trained on an original dataset using real driving traces and a malicious dataset generated from a reinforcement learning agent. The Reinforcement Learning (RL) agent is trained to create intelligent and stealthy attacks that can evade simple detection rules while also giving a malicious EV high charging priority. We evaluate the effectiveness of the generated attacks compared to handcrafted attacks. Moreover, our detection model trained with RL-generated attacks displays greater robustness to intelligent and stealthy attacks.