Trajectory Representation Learning on Grids and Road Networks with Spatio-Temporal Dynamics

Many applications, e.g., Uber, collect large-scale trajectory data from moving vehicles on road network. Trajectory data analytics can benefit many real-world applications, such as route planning and transportation optimizations. Two core operations in trajectory data analytics are trajectory similarity search and join, and both of them rely on a trajectory similarity function to measure the similarity between two trajectories. However, existing similarity functions focus on trajectory points distance and neglect the fact the trajectories should be on road network. Obviously aligning trajectories on road network can remove the noise points introduced by system errors. Toward this goal, we define a road-network-aware trajectory similarity function to measure trajectory similarity. To support trajectory similarity search and join, we propose a filtering-refine framework. In the filtering step, we compute a signature of each trajectory such that if two trajectories are similar, they must share a common signature. We utilize the signatures to prune a huge number of dissimilar pairs. In the refine step, we design effective algorithms to verify the candidates that are not pruned in the filtering step. To support large-scale trajectories, we develop a system DISON for Distributed In-Memory Trajectory Similarity Search and Join on Road Network. DISON splits trajectories into disjoint partitions by considering load balance and locality, and designs effective global index to prune irrelevant partitions. Extensive experiments on real datasets showed that our method achieved high effectiveness, efficiency, and scalability and outperformed existing solutions significantly.

In contrast to conventional digital computers that operate as instructed by programmers, biological organisms solve problems and make decisions through intrinsic spatiotemporal dynamics in which their dynamic components process environmental information in a self-organized manner. Previously, we formulated two mathematical models of spatiotemporal dynamics by which the single-celled amoeba (a plasmodial slime mold), which exhibits complex spatiotemporal oscillatory dynamics and sophisticated computing capabilities, could solve a problem and make a decision by changing its amorphous shape in dynamic and uncertain environments. These models can also be implemented by various physical systems that exhibit suitable spatiotemporal dynamics resembling the amoeba’s shape-changing capability. Here we demonstrate that the photoexcitation transfer phenomena in certain quantum nanostructures mediated by optical near-field interactions mimic the amoeba-like spatiotemporal dynamics and can be used to solve two highly complex problems; the satisfiability problem, which is one of the most difficult combinatorial optimization problems, to determine whether a given logical proposition is self-consistent, and the multi-armed bandit problem, which is a decision-making problem in finding the most profitable option from among a number of options that provide rewards with different unknown probabilities. Our problem-solving and decision-making models exhibited better performances than conventionally known best algorithms. These demonstrations pave the way for a novel nanophotonic computing paradigm in which both coherent and dissipative processes are exploited for performing powerful solution searching and efficient decision-making with low energy consumption.KeywordsSlot MachineLower Energy LevelSatisfying AssignmentRadiation ProbabilityBandit ProblemThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Urban congestion has become a global problem with urbanization and motorization. The analysis of time-varying traffic congestion patterns is necessary to formulate effective management strategies. The existing studies have focused on traffic flow patterns developed by the volume, speed and density of road sections in a limited district, while the long-time analysis of congestion patterns of the macro road network at the city level is inadequate. This paper aims to recognize traffic congestion patterns of the urban road network based on the traffic performance index (TPI) of 699 days in 2018, 2019 and 2021 in Beijing. The self-organizing maps (SOM) method improved by an automatic clustering number determination algorithm is proposed to cluster congestion patterns based on time-varying TPI. The traffic congestion of the macro road network is clustered into Mondays, Fridays, ordinary weekdays, congested weekdays, weekdays of winter and summer vacation, Saturdays, Sundays and festivals patterns. Patterns of Mondays and congested weekdays have a prominent morning peak, while patterns of Fridays, ordinary weekdays, and weekdays of winter and summer vacation have a prominent evening peak. Saturdays, Sundays and festivals are less congested than weekday patterns. It is verified that the SOM method proposed in this paper clusters traffic congestion into more detailed and accurate patterns, and it is applicable to TPI clustering in different years. The degree of congestion in 2021 increases by 7.15% in peak hours and decreases by 7.50% in off-peak hours compared with that in 2019 due to COVID-19. This method is helpful for traffic management in terms of making decisions according to different congestion patterns.

Estimated time of arrival (ETA) prediction, also known as travel time estimation, is a fundamental task for a wide range of intelligent transportation applications, such as navigation, route planning, and ride-hailing services. To accurately predict the travel time of a route, it is essential to take into account both contextual and predictive factors, such as spatial-temporal interaction, driving behavior, and traffic congestion propagation inference. The ETA prediction models previously deployed at Baidu Maps have addressed the factors of spatial-temporal interaction (ConSTGAT) and driving behavior (SSML). In this work, we focus on modeling traffic congestion propagation patterns to improve ETA performance. Traffic congestion propagation pattern modeling is challenging, and it requires accounting for impact regions over time and cumulative effect of delay variations over time caused by traffic events on the road network. In this paper, we present a practical industrial-grade ETA prediction framework named DuETA. Specifically, we construct a congestion-sensitive graph based on the correlations of traffic patterns, and we develop a route-aware graph transformer to directly learn the long-distance correlations of the road segments. This design enables DuETA to capture the interactions between the road segment pairs that are spatially distant but highly correlated with traffic conditions. Extensive experiments are conducted on large-scale, real-world datasets collected from Baidu Maps. Experimental results show that ETA prediction can significantly benefit from the learned traffic congestion propagation patterns. In addition, DuETA has already been deployed in production at Baidu Maps, serving billions of requests every day. This demonstrates that DuETA is an industrial-grade and robust solution for large-scale ETA prediction services.

The pervasiveness of GPS-enabled devices and wireless communication technologies results in massive trajectory data, incurring expensive cost for storage, transmission, and query processing. To relieve this problem, in this paper we propose a novel framework for compressing trajectory data, REST ( Re ference-based S patio- t emporal trajectory compression), by which a raw trajectory is represented by concatenation of a series of historical (sub-)trajectories (called reference trajectories) that form the compressed trajectory within a given spatio-temporal deviation threshold. In order to construct a reference trajectory set that can most benefit the subsequent compression, we propose three kinds of techniques to select reference trajectories wisely from a large dataset such that the resulting reference set is more compact yet covering most footprints of trajectories in the area of interest. To address the computational issue caused by the large number of combinations of reference trajectories that may exist for resembling a given trajectory, we propose efficient greedy algorithms that run in the blink of an eye and dynamic programming algorithms that can achieve the optimal compression ratio. Compared to existing work on trajectory compression, our framework has few assumptions about data such as moving within a road network or moving with constant direction and speed, and better compression performance with fairly small spatio-temporal loss. In addition, by indexing the reference trajectories directly with an in-memory R-tree and building connections to the raw trajectories with inverted index, we develop an extremely efficient algorithm that can answer spatio-temporal range queries over trajectories in their compressed form. Extensive experiments on a real taxi trajectory dataset demonstrate the superiority of our framework over existing representative approaches in terms of both compression ratio and efficiency.

Nowadays, most vehicles are equipped with positioning devices such as GPS which can generate a tremendous amount of trajectory data and upload them to the server in real time. The trajectory data can reveal the shape and evolution of the road network and therefore has an important value for road planning, vehicle navigation, traffic analysis, and so on. In this paper, a road network generation method is proposed based on the incremental learning of vehicle trajectories. Firstly, the input vehicle trajectory data are cleaned by a preprocess module. Then, the original scattered positions are clustered and mapped to the representation points which stand for the feature points of the real roads. After that, the corresponding representation points are connected based on the original connection information of the trajectories. Finally, all representation points are connected by a Delaunay triangulation network and the real road segments are found by a shortest path searching approach between the connected representation point pairs. Experiments show that this method can build the road network from scratch and refine it with the input data continuously. Both the accuracy and timeliness of the extracted road network can continuously be improved with the growth of real-time trajectory data.

Brain function is understood to be regulated by complex spatiotemporal dynamics, and can be characterized by a combination of observed brain response patterns in time and space. Magnetoencephalography (MEG), with its high temporal resolution, and functional magnetic resonance imaging (fMRI), with its high spatial resolution, are complementary imaging techniques with great potential to reveal information about spatiotemporal brain dynamics. Hence, the complementary nature of these imaging techniques holds much promise to study brain function in time and space, especially when the two data types are allowed to fully interact. We employed coupled tensor/matrix factorization (CMTF) to extract joint latent components in the form of unique spatiotemporal brain patterns that can be used to study brain development and function on a millisecond scale. Using the CMTF model, we extracted distinct brain patterns that revealed fine-grained spatiotemporal brain dynamics and typical sensory processing pathways informative of high-level cognitive functions in healthy adolescents. The components extracted from multimodal tensor fusion possessed better discriminative ability between high- and low-performance subjects than single-modality data-driven models. Multimodal tensor fusion successfully identified spatiotemporal brain dynamics of brain function and produced unique components with high discriminatory power. The CMTF model is a promising tool for high-order, multimodal data fusion that exploits the functional resolution of MEG and fMRI, and provides a comprehensive picture of the developing brain in time and space.

Modern navigation systems and logistics services rely heavily on digital road networks, which can be accurately constructed from crowd-sourced GPS trajectories. Yet, most existing trajectory-based road network generators project GPS coordinates to 2D grid, followed by semantic segmentation to obtain road heatmap, and further processed to road network graph via heuristic processing. The data through this pipeline involve multiple cross-modal heuristic conversions, which cause serious information loss and harm overall robustness. In this paper, we propose a novel end-to-end learnable deep learning approach, named GraphWalker, to generate accurate road network graphs directly from raw trajectories with minimal information loss. In particular, GraphWalker is a latent diffusion model (LDM) consisting of two key components: the <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Trajectory-to-Walks Diffusion Transformer</b> (T2W-DiT) generates <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">walks</i> with trajectories as conditions, and the <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Walks-to-Graph Variational Auto-Encoder</b> (W2G-VAE) reconstructs road network graphs from walks. GraphWalker keeps the data in geographical space throughout the pipeline to retain all information within raw trajectories. Additionally, the end-to-end learnable design removes the heuristic processing and enhances the robustness and generalizability. Extensive experiments on multiple datasets demonstrate that the proposed GraphWalker can effectively generate high-quality road networks from noisy and sparse trajectories, showcasing significant improvements of up to 55.43% over state-of-the-art.

The market for on-demand mobility services is growing worldwide. These services include, for example, ride-hailing, ride-sharing, and car-sharing. Large-scale fleets of such services collect GPS trajectory (probe vehicle) data constantly everywhere in the network. At a certain penetration rate, this data becomes representative of the entire road network. It can give valuable insights into traffic dynamics and the evolution of congestion. In this paper, we use such GPS trajectory data from Chengdu, China, to investigate the stability and recurrence of macroscopic traffic patterns. Using the two-fluid theory, we find that the two-fluid coefficients are robust on between-day variation, not only supporting the theory itself but also emphasizing that the general evolution of traffic is a robust pattern. We investigate the deviations from the model using time series analysis of the residuals of the two-fluid model. Here, we find evidence for daily and weekly seasonality in the residuals, indicating that congestion patterns are convincingly recurring. These patterns can be used for network-wide traffic state prediction. We conclude that GPS trajectory data from large on-demand mobility fleets is a promising data source for observing traffic patterns in urban road networks once the data becomes representative.

To improve the efficiency of detecting abnormal traffic incidents on the road network and reduce the false alarm rate, a real-time traffic anomaly detection framework based on a graph spatiotemporal pattern learning (GSTPL) network is proposed. In this framework, a traffic pattern search algorithm based on a fluctuation similarity measure is designed to screen traffic flow data with the same traffic pattern, and a traffic pattern graph tuple is constructed as the input of the network model to avoid the sample imbalance problem and the effect of single-sample randomness for traffic pattern learning. Then the GSTPL network is designed to extract, unsupervised, the traffic spatiotemporal pattern features and make a reasonable prediction of future traffic parameters as the basis for anomaly evaluation. To further restrain the effect of random fluctuations in traffic flow parameters, an abnormal state evaluation method is designed to calculate the anomaly state likelihood by prediction error distribution learning. The overall detection framework realizes stable prediction of network key node traffic parameters by using spatiotemporal pattern features to construct the traffic pattern graph tuple, and gives incident evaluation results in real time by combination with the detection data. The experiment uses I90 and I405 highway traffic data in Seattle, WA, from 2015. Through comparative analysis, the proposed incident detection method based on GSTPL has a higher detection rate and lower false alarm rate, can adaptively learn dynamic changes of the traffic pattern, and has strong adaptability and stability to different traffic environments.

In a cloud environment, resources should be acquired quickly and automatically released at runtime. Traditional trajectory data partitions, indexing and query processing techniques are extended, so that they can take advantage of the cloud of large clusters of highly parallel processing capabilities. There are ways with trajectory data in the cloud database query processing. The advanced sensing techniques available today caused the existence of many types of trajectory datasets whose study can form a strong basis for decision making regarding a data context. Trajectory data generally includes trajectory classification, trajectory clustering, and trajectory associations and so on. All these tools need trajectory similarity measures to get comparisons in trajectory data. Some of the popular existing methods of trajectory similarity include Euclidean distance, semantic distance and there near variants. Each variant of these techniques exhibits notable differences in measurement of similarity and computational difficulty. This paper targeted to deal with this situation with the goal of lower computational efforts in trajectory similarity computations. A two‐phase trajectory similarity measure is defined. The first phase does point level analysis of trajectory points representing all trajectories and the following phase use the analysis to find the similarities. Using a hybrid similarity framework, new means for trajectory clustering and trajectory query processing are developed. From trajectory analysis followed by trajectory clustering, evaluation of cluster quality and processing of trajectory query are undertaken in this paper in different ways from the existing methods. All the proposed developments are tested with trajectory datasets with multiple attributes and the results are more favour to proposed framework than the existing ones.

The GPS traces collected from a large taxi fleet provide researchers novel opportunities to inspect the urban dynamics in a city and lead to applications that can bring great benefits to the public. In this chapter, based on a real life large-scale taxi GPS dataset, the authors reveal the unique characteristics in the four different trace stages according to the passenger status, study the urban dynamics revealed in each stage, and explain the possible applications. Specifically, from passenger vacant traces, they study the taxi service dynamics, introduce how to use them to help taxis and passengers find each other, and reveal the work shifting dynamics in a city. From passenger occupied traces, they introduce their capabilities in monitoring and predicting urban traffic and estimating travel time. From the pick-up and drop-off events, the authors show the passenger hotspots and human mobility patterns in a city. They also consider taxis as mobile GPS sensors, which probe the urban road infrastructure dynamics.

Road intersections are important components of urban road system. It is the traffic flow characteristics representing the current traffic situation that provide a basis for the planning, designing and management of intersections. In this paper, we constructed an automatic processing framework on traffic flow characteristics analysis and understanding the traffic state at the urban road intersections, based on the collected raw vehicle motion trajectories. Our proposed method is basically attributed to identifying distinct vehicle motion patterns at intersections hierarchically using raw trajectory. Firstly, the fundamental assumption in traditional approaches that the trajectory set of high quality are readily available after manual rectification is not taken for granted any more. And by fully analyzing the local characteristics of trajectories, we figure out and explain various patterns behind traffic flow as well as yielded higher accuracy in motion trajectory clustering under the multi-layer spectral clustering method. At last, coupling the analyzing results with the surrounding characteristics of the intersection, the examples computing traffic flow features and predicting vehicle activity illustrates the potential of applying vehicle trajectories to traffic study, which are all suggested by experimental results.

All-optical schemes for controlling complex delay-induced and spatio-temporal dynamics in multi-stripe and broad-area semiconductor lasers are presented. The underlying control-principle is based on continuous delayed optical feedback. Its influence on the general spatiotemporal dynamics of semiconductor lasers is discussed. The delayed feedback control principle allows successful control of delay-induced chaotic behavior as well as the dynamic optical filamentation in broad-area and multi-stripe lasers. The suppression of spatiotemporal instabilities in semiconductor laser arrays is achieved by selectively stabilizing relevant transverse optical modes. The relevant values of delay-times and delay-phase are obtained from a nonlinear eigenmode analysis based on a complex extension of the Karhunen-Loeve decomposition. Thereby a destructive interference condition in the higher-order transverse modes conveying the instabilities is achieved. The stabilization technique is demonstrated on multi-stripe laser arrays. In high-power broad-area lasers the spatiotemporal filamentation processes occur on both macroscopic and microscopic scales. Numerical simulations on the basis of Maxwell-Bloch equations for spatially extended semiconductor lasers reveal the internal spatial and temporal scales on which control consequently has to occur. It is demonstrated that the combined filamentation and propagation effects may be controlled by suitable spatially structured delayed optical feedback allowing, in particular, the control of coherent regimes in originally temporally and spatially chaotic states.

Road intersection is an important part of the road network, and it plays a key role in many intelligent transportation applications. Currently, the generation of road maps required expensive field surveying and labor-intensive mapping. With more and more public vehicles equipped with positioning devices, massive trajectory data are available. These trajectory data provides a new way to generate and update urban road networks in real time. Although numerous methods are currently available to reconstruct road maps from GPS trajectories, most of these methods focus on the generation of road segments (i.e., road centerlines or road lanes). Road intersections are simply represented by nodes, the generation of road intersections with detailed structures still remains a challenge. Therefore, this paper proposes a novel method for constructing road intersections from vehicle GPS trajectories. First, the boundaries of road intersections are detected by turning angles clustering. Then, the entrances and exits of each intersection are identified based on a newly developed method. Finally, the detailed geometric structures of intersections are reconstructed based on the detected entrances and exits, and turning rules are further extracted and assigned to each intersection. The experiments on real-world trajectory data show that the proposed method can generate the detailed geometric structures of road intersections as well as the turning rule semantic information successfully.