Understanding the spatial dynamics of bike-sharing usage is critical for effective urban planning and mobility resource management. In this study, we propose an interpretable deep learning approach to uncover spatial relationships embedded in bike-sharing activities. Specifically, we develop a spatially-adapted SHapley Additive exPlanations (SHAP)-based method to quantify the spatial dependencies between locations in bike-sharing activities and apply it to interpret the predictions of a bike-sharing model. Extensive experiments upon Citi Bike data from New York City in December 2023 reveal that spatial influence does not strictly follow geographic proximity and is anisotropic. Additionally, non-member users exhibit weaker spatial dependencies in their bike usage behavior, resulting in lower short-term predictability compared to member users. Our studies shed deep insights into the spatial dynamics of bike-sharing systems and provide guidance for more effective service deployment and system design.

In the complex traffic environments, understanding how a focal vehicle interacts (e.g., maneuvers) with various traffic elements (e.g., other vehicles, pedestrians, and road infrastructures), i.e., vehicle-to-X interactions (VXIs), is essential for developing the advanced driving support and intelligent vehicles. To derive the VXI scene understanding, reasoning, and decision support (e.g., suggesting cautious move in response of a pedestrian crossing the street), this work takes into account the recent advances of multi-modality large language models (MLLMs). We develop VXI-SUR, a novel VXIScene Understanding and Reasoning system based on vision-language modeling. VXI-SUR takes in the visual VXI scene, and generates the structured textual responses that interpret the VXI scene and suggests an appropriate decision (e.g., braking, slowing down). We have designed within VXI-SUR a VXI memory mechanism with both scene and knowledge augmentation mechanisms, and enabled scene-knowledge co-learning to capture complex correspondences across scenes and decisions. We have performed extensive and comprehensive evaluations of VXI-SUR based on an open-source dataset with ∼17k VXI scenes. We have conducted extensive experimentation studies upon VXI-SUR, and corroborated VXI awareness, description preciseness, semantic matching, and quality in understanding and reasoning the complex VXI scenes.

We propose an equity-awareGRAph-fusion differentiablePooling neural network to accurately predict the spatio-temporal urban mobility (e.g., station-level bike usage in terms of departures and arrivals) withEquity (GRAPE).GRAPEconsists of two independent hierarchical graph neural networks for two mobility systems—one as a target graph (i.e., a bike sharing system) and the other as an auxiliary graph (e.g., a taxi system). We have designed a convolutional fusion mechanism to jointly fuse the target and auxiliary graph embeddings and extract the shared spatial and temporal mobility patterns within the embeddings to enhance prediction accuracy. To further improve the equity of bike sharing systems for diverse communities, we focus on the bike resource allocation and model prediction performance, and propose to regularize the predicted bike resource as well as the accuracy across advantaged and disadvantaged communities, and thus mitigate the potential unfairness in the predicted bike sharing usage. Our evaluation of over 23 million bike rides and 100 million taxi trips in New York City and Chicago has demonstratedGRAPEto outperform all of the baseline approaches in terms of prediction accuracy (by 15.80% for NYC and 50.55% for Chicago on average) and social equity awareness (by 32.44% and 24.43% in terms of resource fairness for NYC and Chicago, and 13.36% and 16.52% in terms of performance fairness).

Accurate prediction of citywide crowd activity levels (CALs), i.e., the numbers of participants of citywide crowd activities under different venue categories at certain time and locations, is essential for the city management, the personal service applications, and the entrepreneurs in commercial strategic planning. Existing studies have not thoroughly taken into account the complex spatial and temporal interactions among different categories of CALs and their extreme occurrences, leading to lowered adaptivity and accuracy of their models. To address above concerns, we have proposed IE-CALP , a novel spatio-temporal I nteractive attention-based and E xtreme-aware model for C rowd A ctivity L evel P rediction. The tasks of IE-CALP consist of (a) forecasting the spatial distributions of various CALs at different city regions (spatial CALs), and (b) predicting the number of participants per category of the CALs (categorical CALs). To realize above, we have designed a novel spatial CAL-POI interaction-attentive learning component in IE-CALP to model the spatial interactions across different CAL categories, as well as those among the spatial urban regions and CALs. In addition, IE-CALP incorporate the multi-level trends (e.g., daily and weekly levels of temporal granularity) of CALs through a multi-level temporal feature learning component. Furthermore, to enhance the model adaptivity to extreme CALs (e.g., during extreme urban events or weather conditions), we further take into account the extreme value theory and model the impacts of historical CALs upon the occurrences of extreme CALs. Extensive experiments upon a total of 738,715 CAL records and 246,660 POIs in New York City (NYC), Los Angeles (LA), and Tokyo have further validated the accuracy, adaptivity, and effectiveness of IE-CALP ’s interaction-attentive and extreme-aware CAL predictions.

On-demand delivery has experienced rapid growth in recent years, revolutionizing people's lifestyles with its timeliness and convenience. The order dispatch process in on-demand delivery isconcurrent, wherein couriers continuously accept new orders and deliver them to customers within strict time constraints and dynamic demand and supply. Most of the existing order dispatch mechanisms are designed for independent dispatch or concurrent dispatch without strict deadlines, rendering them unsuitable for real-time concurrent dispatch in on-demand delivery. To address the challenge, we propose aTime-ConstrainedActor-Critic Reinforcement learning based concurrent dispatch system calledTCAC-Dispatchto reduce the overdue rate and enhance the long-term revenue. Specifically, we first design a deep matching network (DMN) with a variable action space, which integrates both states embedding (including route behaviors encoding) and actions' embedding into a long-term value for dispatching decisions. Additionally, we design a time-constrained action pruning module to ensure compliance with time constraints. Then we utilize the Actor-Critic framework to tackle the concurrent dispatch considering strict time constraints and stochastic demand-supply. To further optimize the efficiency and delivery resource utilization, we propose an extension of TCAC (i.e., TCAC+), which consists of (i) a learning-based order service time prediction module to determine whether to relax the deadline of some orders; and (ii) a multi-critic framework to optimize concurrent order dispatch with both tight deadlines and relaxed deadlines using dynamic weighting mechanism. We evaluate the TCAC-Dispatch with one-month data involved with 36.48 million orders and 42,000 couriers collected from Eleme, one of the largest on-demand delivery companies in China. Experiments are conducted on a data-driven emulator deployed on the development environment of Eleme and the results demonstrate that our method outperforms state-of-the-art baselines with various metrics in both tight deadline and mixed deadline scenarios.

Thanks to advances in the urban big data, the bike sharing, especially station-based bike sharing, has emerged as the important first-/last-mile connectivities in many smart cities. Bike station network (BSN) expansion recommendation, i.e., recommending placement locations of new stations, is essential for satisfying local mobility demands, enhancing the BSN service quality, and may significantly affect the resource fairness and accessibility of different communities in the neighborhood. Furthermore, the dynamic and complex urban mobility environments make the station placement highly challenging to satisfy the mobility needs.

The occurrence of special contexts or events (e.g., extreme weather conditions, festival events, other urban anomalies) can significantly influence the movement patterns of urban mobility (e.g., human crowds, transportation systems). Accurate mobility modeling and prediction under the occurrences of such anomaly events is therefore imperative for city management and urban resource allocation. In this study, we propose EALGAP, a novel Extreme-Aware Local-Global Attention urban mobility Prediction model. Specifically, EALGAP models the spatio-temporal global and local impacts of mobility in different regions and time steps for mobility prediction at various city regions. EALGAP takes into account the global impacts by extracting the overall or regular spatial dependencies and temporal patterns of mobility systems for different regions. We have designed a temporally-varying normalization and data-driven technique to quantify the extreme degrees, i.e., how significantly the extreme events have impacted the local mobility trend, of the patterns within different regions and time steps. We have conducted ex-tensive experimental studies upon four different mobility datasets (over 13 million trips in total) harvested from two metropolitan cities in U.S. with anomalous natural or social events (e.g., hurricane events, other extreme weather conditions, and the Federal holidays). Our results have demonstrated the accuracy, effectiveness, and extreme-awareness of our proposed EALGAP with more than 44.12% error reduction on average compared with other state-of-the-art approaches.

Understanding and learning the actor-to-X inter-actions (AXIs), such as those between the focal vehicles (actor) and other traffic participants (e.g., other vehicles, pedestrians) as well as traffic environments (e.g., city/road map), is essential for the development of a decision-making model and simulation of autonomous driving (AD). Existing practices on imitation learning (IL) for AD simulation, despite the advances in the model learnability, have not accounted for fusing and differentiating the heterogeneous AXIs in complex road environments. Furthermore, how to further explain the hierarchical structures within the complex AXIs remains largely under-explored. To overcome these challenges, we propose HGIL, an interaction- aware and hierarchically-explainable Heterogeneous _Graph- based Imitation Learning approach for AD simulation. We have designed a novel heterogeneous interaction graph (HIG) to provide local and global representation as well as awareness of the AXIs. Integrating the HIG as the state embeddings, we have designed a hierarchically-explainable generative adversarial imitation learning approach, with local sub-graph and global cross-graph attention, to capture the interaction behaviors and driving decision-making processes. Our data-driven simulation and explanation studies have corroborated the accuracy and explainability of HGIL in learning and capturing the complex AXIs.

Thanks to the increasing number and massive coverage, delivery drones, equipped with various sensors, have demonstrated significant but unexplored potentials for large-scale and low-cost urban sensing during package delivery. In this paper, we propose novel studies on the reutilization of such delivery drone resources to fill this void in urban crowdsensing. Accounting for interdependency between flying/sensing and drone delivery weight, we jointly optimize route selection, sensing time, and delivery weight allocation, to maximize delivery and sensing utility under drones’ energy constraints. This problem is formulated as a non-convex mixed-integer non-Linear programming problem, which is proved to be NP-hard. To address this intricate problem, we propose near-optimal algorithms that leverage the equivalent objective function construction, the local search scheme, and the alternating iteration technique. Theoretical analysis indicates that our algorithms can achieve$\frac{1}{4+\varepsilon }$-approximation ratio (where$\varepsilon$is an arbitrarily small positive parameter) and the convergence guarantee in polynomial time, for the scenarios of fixed and adjustable delivery weights, respectively. Extensive trace-based simulations, field experiments, and the real-world application demonstrate that ours can significantly improve the delivery & sensing utility by$124.7\%$and the energy utilization rate by$72.2\%$on average, compared with the drone delivery without reusing.

Of all indoor localization techniques, vision-based localization emerges as a promising one, mainly due to the ubiquity of rich visual features. Visual landmarks, which present distinguishing textures, play a fundamental role in visual indoor localization. However, few researches focus on visual landmark labeling. Preliminary arts usually designate a surveyor to select and record visual landmarks, which is tedious and time-consuming. Furthermore, due to structural changes (e.g., renovation), the visual landmark database may be outdated, leading to degraded localization accuracy. To overcome these limitations, we propose VILL , a user-friendly, efficient, and accurate approach for visual landmark labeling. VILL asks a user to sweep the camera to take a video clip of his/her surroundings. In the construction stage, VILL identifies unlabeled visual landmarks from videos adaptively according to the graph-based visual correlation representation. Based on the spatial correlations with selected anchor landmarks, VILL estimates locations of unlabeled ones on the floorplan accurately. In the update stage, VILL formulates an alteration identification model based on the judgments from different users to identify altered landmarks accurately. Extensive experimental results in two different trial sites show that VILL reduces the site survey substantially (by at least 65.9%) and achieves comparable accuracy.

With the popularity of Mobility-on-Demand (MOD) vehicles, a new market called MOD-Vehicular-Crowdsensing (MOVE-CS) was introduced for drivers to earn more by collecting road data. Unfortunately, MOVE-CS failed after two years of operation. To identify the root cause, we survey 581 drivers and reveal its simple operation model based on blindly competitive rewards. This model brings most drivers few yields, resulting in their withdrawals. In contrast, a similar market termed MOD-Human-Crowdsensing (MOMAN-CS) remains successful thanks to a complex model based on exclusively customized rewards. Hence, we wonder whether MOVE-CS can be resurrected by learning from MOMAN-CS. Despite considerable similarity, we can hardly apply the operation model of MOMAN-CS to MOVE-CS, since drivers are also concerned with passenger missions that dominate their earnings. To this end, we analyze a large-scale dataset of 12,493 MOD vehicles, finding that drivers have explicit preference for short-term, immediate gains as well as implicit rationality in pursuit of long-term, stable profits. Therefore, we design a novel operation model for MOVE-CS, at the heart of which lies a spatial-temporal differentiation-aware task recommendation scheme empowered by submodular optimization. Applied to the dataset, our design would essentially benefit both the drivers and platform, thus possessing the potential to resurrect MOVE-CS.

Urban dockless e-scooter sharing (DES) has become a popular Web-of-Things (WoT) service and widely adopted globally. Despite its early commercial success, conventional mobility demand and supply prediction based on machine learning and subsequent redistribution may favor advantaged socio-economic communities and tourist regions, at the expense of reducing mobility accessibility and resource allocation for historically disadvantaged communities. To address this unfairness, we propose a socially-Equitable Interactive Graph information fusion-based mobility flow prediction system for Dockless E-scooter Sharing (EIGDES). By considering city regions as nodes connected by trips, EIGDES learns and captures the complex interactions across spatial and temporal graph features through a novel interactive graph information dissemination and fusion structure. We further design a novel model learning objective with metrics that capture both the mobility distributions and the socio-economic factors, ensuring spatial fairness in the communities’ resource accessibility and their experienced DES prediction accuracy. Through its integration with the optimization regularizer, EIGDES jointly learns the DES flow patterns and socio-economic factors, and returns socially-equitable flow predictions. Our in-depth experimental study upon more than 2,122,270 DES trips from three metropolitan cities in North America has demonstrated EIGDES’s effectiveness in accurate prediction of DES flow patterns with substantial reduction of mobility unfairness.

Falls are the second leading cause of accidental or unintentional injuries/deaths worldwide. Accurate pose estimation using commodity mobile devices will help early detection and injury assessment of falls, which are essential for the first aid of elderly falls. By following the definition of fall, we propose a P ervasive P ose Est imation scheme for fall detection ( P \( ^2 \) Est ), which measures changes in tilt angle and height of the human body. For the tilt measurement, P \( ^2 \) Est leverages the pointing of the mobile device, e.g., the smartphone, when unlocking to associate the Device coordinate system with the World coordinate system. For the height measurement, P \( ^2 \) Est exploits the fact that the person’s height remains unchanged while walking to calibrate the pressure difference between the device and the floor. We have prototyped and tested P \( ^2 \) Est in various situations and environments. Our extensive experimental results have demonstrated that P \( ^2 \) Est can track the body orientation irrespective of which pocket the phone is placed in. More importantly, it enables the phone’s barometer to detect falls in various environments with decimeter-level accuracy.

Bike sharing service (BSS) networks have been proliferating all over the globe thanks to their success as the first/last-mile connectivity inside a smart city. Their (re)configuration — i.e., station (re)placement and dock resizing — has thus become increasingly important for BSS providers and smart city planners. Instead of using conventional labor-intensive manual surveys, we propose a novel information fusion framework calledCBikesthat (re)configures the BSS network by jointly fusing crowdsourced station suggestions from online websites and the usage history of bike stations. Using comprehensive real data analyses, we identify and exploit important global trip patterns to (re)configure the BSS network while mitigating the local biases of individual feedbacks. Specifically, crowdsourced feedbacks, station usage, cost and other constraints are fused into a joint optimization of BSS network configuration. We also model the spatial distributions of station usage to account for and estimate the unexplored regions without historical usage information. We further design a semidefinite programming transformation to solve the bike station (re)placement problem efficiently and effectively. Our extensive data analytics and evaluation have shownCBikes’ effectiveness and accuracy in (re)placing stations and resizing docks based on three large BSS systems (with$>$900 stations) in Chicago, Twin Cities (Minneapolis–Saint Paul), and Los Angeles.

As an alternative means of convenient and smart transportation, mobility-on-demand (MOD), typified by online ride-sharing and connected taxicabs, has been rapidly growing and spreading worldwide. The large volume of complex traffic and the uncertainty of market supplies/demands have made it essential for many MOD service providers toproactivelydispatch vehicles towards ride-seekers. To meet this need effectively, we proposeSTRide, an MOD coordination learning mechanism reinforced spatio-temporally with capsules. We formalize the adaptive coordination of vehicles into a reinforcement learning framework.STRideincorporates spatial and temporal distributions of supplies (vehicles) and demands (ride requests), customers’ preferences and other external factors. A novel spatio-temporal capsule neural network is designed to predict the provider’s rewards based on MOD network states, vehicles and their dispatch actions. This way, the MOD platform adapts itself to the supply-demand dynamics with the best potential rewards. We have conducted extensive data analytics and experimental evaluation with five large-scale datasets ($\sim$27 million rides from Uber, NYC/Chicago Taxis, Didi and Car2Go).STRideis shown to outperform state-of-the-arts, substantially reducing request-rejection rate and passenger waiting time, and also increasing the service provider’s profits.

Dockless E-scooter Sharing (DES) has become a popular means of last-mile commute for many smart cities. As e-scooters are getting deployed dynamically and flexibly across city regions that expand and/or shrink, accurate prediction of the e-scooter distribution given the reconfigured regions becomes essential for city planning. We presentGCScoot, a novel flow distribution prediction approach for reconfiguring urban DES systems. Based on real-world datasets with reconfiguration, we analyze e-scooter distribution features and flow dynamics for the data-driven designs. We propose a novel spatio-temporal graph capsule neural network withinGCScootto predict future dockless e-scooter flows given the reconfigured regions.GCScootpre-processes historical spatial e-scooter distributions into flow graph structures, where discretized city regions are considered as nodes and inter-region flows as edges. To facilitate initial training, we cluster the regions and generate virtual data for new deployment regions based on their peers in the same cluster. Given above designs, the region-to-region correlations embedded within the temporal flow graphs are captured via the multi-graph capsule convolutional neural network which accurately predicts the DES flows. Extensive studies upon four e-scooter datasets (total$>$3.4 million rides) in four populous US cities have corroborated accuracy and effectiveness ofGCScootin predicting the e-scooter distributions.

Taking into account the availability of the historical GPS trajectories of drivers, given a new GPS trajectory, Driver mobility fingerprint (DMF) identification aims at (i) determining whether a generated trajectory belongs to a potential driver, and (ii) detecting if a trajectory is likely anomalous based on a driver's historical data. Prior studies often consider hand-crafted feature engineering techniques to extract DMFs while contextual factors like weather and points-of-interest (POIs) are hardly accounted for, which might not achieve satisfactory identification results. To address above, we propose RM-Drive, a novel framework based on reinforced feature extraction and multi-resolution learning. Specifically, we first employ spatio-temporal inverse reinforcement learning (ST-IRL) to extract DMFs from historical trajectories. Then, we generate trajectory embeddings by fusing the extracted DMFs and the contextual factors using the multi-resolution trajectory embedding network (MTE-Net). Our proposed MTE-Net consists of multi-resolution convolutional neural network (MR-CNN), which enables the model to learn the multi-resolution features of the DMFs. Finally, we leverage the trajectory embeddings for the driver classification and anomaly detection. We have conducted extensive evaluation studies upon RM-Drive with two real-world datasets, and our results demonstrate the performance improvements from the state-of-the-art of driver classification and anomaly detection respectively by 21% and 11% on average based on several evaluation metrics, including accuracy, precision, and recall, etc.

Instant delivery has developed rapidly in recent years and significantly changed the lifestyle of people due to its timeliness and convenience. In instant delivery, the order dispatch process is concurrent. Couriers take new orders continuously and deliver multiple orders in a delivery trip (i.e., a batch). The delivery time of orders in a batch is often overlapped and interlinked with each other. The pickup and delivery sequence of the existing orders in a batch changes dynamically due to time constraints and real-time overdue possibility (i.e., the rate of deliveries that are not finished in promised time). Most of existing order dispatch mechanisms are designed for independent order dispatch or concurrent delivery without strict time constraints, hence are incapable of handling real-time concurrent dispatch with strict time constraints in on-demand instant delivery. To address the challenge, we propose a Time-Constrained Actor- Critic Reinforcement learning based concurrent dispatch system called TCAC-Dispatch to enhance the long-term overall revenue and reduce the overdue rate. Specifically, we design a deep matching network (DMN) with a variable action space, which integrates the state embedding (including route behaviors encoding) and actions embedding features into a long-term matching value. Then the Actor-Critic model tackles the concurrent order dispatch problem considering strict time constraints and stochastic demand-supply in instant delivery. An estimated time-based action pruning module is designed to ensure time constraints guarantee and accelerate the training as well as dispatching processes. We evaluate the TCAC-Dispatch with one-month data involved with 36.48 million orders and 42,000 couriers collected from one of the largest instant delivery companies in China, i.e., Eleme. Empirical experiments are conducted on a data-driven emulator deployed on the development environment of Eleme and results show that our method achieves 22% of the increase in total revenue and reduces the overdue rate by 21.6%.

For effective crowdsensing, it is essential to incentivize the interactions of participants and platforms. Existing approaches do not tailor users’ bidding to their preferences, i.e., personalized bidding (PB). To meet this need, we design an incentive mechanism, called Picasso, that achieves not only the expressiveness and description efficiency of PB for users, but also minimal social cost, computational efficiency, and strategy proof for platform owners. This design is, however, challenging due to the intrinsic conflicting goals of the platform owner and users. To handle these conflicts, Picasso represents bids in a novel 3-D expression space by orchestrating three logical operations to balance among expressiveness, computational complexity, and description efficiency. Moreover, we equivalently decompose and recombine the complex task dependencies of bids originated from the expressiveness of PB, thus achieving a constant-factor approximation of optimal task allocation with strategy proof in polynomial time. These properties of Picasso are proven theoretically. In addition to a detailed simulation study, our trace-driven evaluations show that, compared to existing approaches, Picasso can enable each user to bid$9.7\times $more tasks, on average, and decrease the description length by 74%, thus encouraging more users’ participation. Picasso also reduces the platform owner’s payment by more than 61%, hence yielding a win–win solution for incentivizing platform–user interactions.

After years of development, bike sharing has been one of the major choices of transportation for urban residents worldwide. However, efficient use of bike sharing resources is challenging due to the unbalanced station-level demands and supplies, which causes the maintenance of the bike sharing systems painstaking. To achieve system efficiency, efforts have been made on accurate prediction of bike traffic (demands/pick-ups and returns/drop-offs). Nonetheless, bike station traffic prediction is difficult due to the spatio-temporal complexity of bike sharing systems. Moreover, such level of prediction over the entire bike sharing systems is also challenging due to the large number of bike stations.To fill this gap, we propose BikeGAAN, a graph adjacency attention neural network to predict station-level bike traffic for entire bike sharing systems. The proposed prediction system consists of a graph convolutional network with an attention mechanism differentiating the spatial correlations between features of bike stations in the system and a long short-term memory network capturing temporal correlations. We have conducted extensive data analysis upon bike usage, weather, points of interest and event data, and derived the graph representation of the bike sharing networks. Through experimental study on over 27 millions trips of bike sharing systems of four metropolitan cities in the U.S., New York City, Chicago, Washington D.C. and Los Angeles, our network design has shown high accuracy in predicting the bike station traffic in the cities, outperforming other baselines and state-of-art models.

As a healthy, efficient and green alternative to motorized urban travel, bike sharing has been increasingly popular, leading to wide deployment and use of bikes instead of cars. Accurate bike-flow prediction at the individual station level is essential for bike sharing service. Due to the spatial and temporal complexities of traffic networks and the lack of data-driven design for bike stations, existing methods cannot predict the fine-grained bike flows to/from each station.

Thanks to recent progresses in mobile payment, IoT, electric motors, batteries and location-based services, Dockless E-scooter Sharing (DES) has become a popular means of last-mile commute for a growing number of (smart) cities. As e-scooters are getting deployed dynamically and flexibly across city regions that expand and/or shrink, with subsequent social, commercial and environmental evaluation, accurate prediction of the distribution of e-scooters given reconfigured regions becomes essential for the city planners and service providers.

Indoor localization has attracted considerable attention lately, due to its large commercial and social values in smart cities. The existing indoor localization approaches mostly rely on fingerprint techniques, and many of those leverage either spatially discrete fingerprints or temporally consecutive ones for localization, which either suffers from large errors due to signal ambiguities or high time overhead with long sequences. To achieve high accuracy with low computational cost, we propose ST-Loc, a deep neural network that extracts features from multiple representations of a single signal sequence for localization, where each representation indicates a corresponding signal structure with underlying feature correlations. Taking geomagnetism as an example, we infer location features from two different representations, e.g., spatial and temporal. In spatial representation, a signal sequence is converted to a signal heatmap, where each pixel corresponds to a spatial location and the value indicates fingerprint. Temporal representation, on the other hand, is a signal sequence with ordered readings, which provides temporal correlations. Using these different representations, we employ convolutional and recurrent networks to extract location features and fuse them to generate more distinguishing features for localization. We have conducted extensive experiments in two different trial sites, a narrow office area and a spacious food plaza. Our experimental results show that ST- Loc achieves more than 43% average localization error reduction compared with state-of-the-art competing schemes in both trial sites. © 2019 IEEE.

To realize an efficient and flexible urban crowd-flow identification, we propose a new scheme called CFid by fusing fine-grained spatio-temporal smartphone signal features. Considering the abundance of ambient WiFi and geomagnetism, we design and validate several of their fine-grained features and similarity metrics to determine if two individuals belong to the same crowd-flow. We formalize a graph stream clustering problem, where co-flow user pairs are opportunistically identified and fed as a dynamic sequence of edges connecting each other. Given such a flexible form, a fast and accurate algorithm processes the edges and identifies the crowd-flows for further upper-level applications, including urban-flow monitoring and shopping-advertisement/recommendation. Using extensive data-driven analytics and 8-month experimental studies upon 50 users (over 2,500 walking traces at 7 different urban sites), we have validated the accuracy (usually >95%), efficiency (only <;5% extra energy-footprint on average than normal usage on mobiles) and flexibility of CFid in identifying large-scale crowd-flows.

As an alternative means of convenient and smart transportation, mobility-on-demand (MOD), typified by online ride-sharing and connected taxicabs, has been rapidly growing and spreading worldwide. The large volume of complex traffic and the uncertainty of market supplies/demands have made it essential for many MOD service providers to proactively dispatch vehicles towards ride-seekers.

Pricing in mobility-on-demand (MOD) networks, such as Uber, Lyft, and connected taxicabs, is done adaptively by leveraging the price responsiveness of drivers (supplies) and passengers (demands) to achieve such goals as maximizing drivers’ incomes, improving riders’ experience, and sustaining platform operation. Existing pricing policies only respond to short-term demand fluctuations without accurate trip forecast and spatial demand-supply balancing, thus mismatching drivers to riders and resulting in loss of profit. We propose CAPrice, a novel adaptive pricing scheme for urban MOD networks. It uses a new spatio-temporal deep capsule network (STCapsNet) that accurately predicts ride demands and driver supplies with vectorized neuron capsules while accounting for comprehensive spatio-temporal and external factors. Given accurate perception of zone-to-zone traffic flows in a city, CAPrice formulates a joint optimization problem by considering spatial equilibrium to balance the platform, providing drivers and riders/passengers with proactive pricing “signals.” We have conducted an extensive experimental evaluation upon over 4.0× 10 8 MOD trips (Uber, Didi Chuxing, and connected taxicabs) in New York City, Beijing, and Chengdu, validating the accuracy, effectiveness, and profitability (often 20% ride prediction accuracy and 30% profit improvements over the state-of-the-arts) of CAPrice in managing urban MOD networks.

Locality classification is an important component to enable location-based services. It entails two sequential queries: 1) whether a target is within the site or not, i.e., inside/outside region decision, and 2) if so, which area in the region the target is located, i.e., area classification. Locality classification is hence more coarse-grained and efficient as compared with pinpointing the exact target location in the region. The classification problem is challenging, because fingerprints may not exist outside the region for training. Furthermore, the target may sample an incomplete RSSI vector due to, say, random signal noise, momentary occlusion, or scanning duration. The algorithm also has to be computationally efficient. We propose INOA, a scalable and practical locality classification overcoming the above challenges. INOA may serve as a plug-in before any fingerprint-based localization, and can be incrementally extended to cover new areas or regions for large-scale deployment. Its preprocessor cherry-picks only those discriminating access points, which greatly enhances computational efficiency and accuracy. By formulating a “one-class” classifier using ensemble learning, INOA accurately decides whether the target is within the region or not. Extensive experimental trials in different sites validate the high efficiency and accuracy of INOA, without the need of full RSSI vectors collected at the target.

Fusing mutual distance information with fingerprints can substantially improve indoor localization accuracy. Such distance information may be spatial (e.g., measurement among users or from installed beaconing devices) or temporal (e.g., via dead-reckoning). Previous approaches on distance-fusion often require deterministic distance measurement, consider fingerprints and distances separately, or are narrowly applicable to some specific sensing technology or scenario. Given the fact that fingerprint and distance measurements are intrinsically random, we propose Maxlifd, an accurate indoor localization framework fusing fingerprints and distances of arbitrary distributions via joint maximum likelihood. Maxlifd is a generic statistical/probabilistic framework applicable to a wide range of sensors (peer-assisted, INS, iBeacon, etc.) and fingerprints (Wi-Fi, RFID, etc.). It achieves low localization errors by a novel optimization formulation jointly considering mutual distances and fingerprint signals. Using generic probabilistic formulation, we further derive the lower bound on localization error for comprehensive performance analysis. We have implemented Maxlifd, and conducted extensive simulation and experimental trials in an international airport and our university campus. Our results show that Maxlifd achieves significantly lower errors than other state-of-the-art schemes (often by more than 30 percent). We experimentally verify that its performance does not depend sensitively on the exact knowledge of the underlying distributions beyond simple Gaussian.

Image-based indoor localization has aroused much interest recently because it requires no infrastructure support. Previous approaches on image-based localization, due to their computation and storage requirements, often process queries at servers. This does not scale well, incurs round-trip delay, and requires constant network connectivity. Many also require users to manually confirm the shortlisted matched landmarks, which is inconvenient, slow, and prone to selection error. To overcome these limitations, we propose a h ighly a utomated (in terms of image confirmation after taking images) i mage-based l ocalization algorithm (HAIL), distributed in mobile devices. HAIL achieves resource efficiency (in terms of storage and processing) by keeping only distinguishing visual features for each landmark, and employing the efficient k-d tree to search for features. It further utilizes motion sensors and map constraints to enhance the localization accuracy without user operation. We have implemented HAIL on Android platforms and conducted extensive experiments in a food plaza and a premium shopping mall. Experimental results show that it achieves much higher localization accuracy (reducing the localization error by more than 20%) and computation efficiency (by more than 40% in time) as compared with the state-of-the-art approaches.

Geomagnetism is promising for indoor localization due to its omnipresence, high stability, and availability of magnetometers in smartphones. Previous works often fuse it with pedometer via particles, which are computationally intensive and make strong user behavior assumptions. To overcome that, we propose Magil, an approach leveraging geo mag netism for i ndoor l ocalization. To our best knowledge, this is the first piece of work using geomagnetism for smartphone localization without the need of pedometer or user walking model. Magil is applicable to any open or complex indoor environment. In the offline phase, Magil collects and stores geomagnetic fingerprints while surveyors walk indoors. In the online phase, it employs a fast algorithm to match the geomagnetic variation of the target with the stored fingerprints. Given closely matched segments, Magil constructs user trajectory efficiently with a modified shortest path formulation by selecting and connecting these matched segments. To further improve accuracy and deployability, we propose MagFi, which extends Mag il by fusing it with Wi- Fi . MagFi further collects opportunistic Wi-Fi RSSI for fingerprint construction. We have implemented both Magil and MagFi and conducted extensive experiments in our campus. Results show that our schemes outperform state-of-the-art schemes by a wide margin (often cutting localization error by 30%).

In recent years, augmented reality (AR) has drawn significant attention in the automotive industry and shown great potential for a variety of driver-assistance applications. Tracking the driver's head is vital to seamlessly merge the AR content in the driver's view but still remains an open problem. Specifically, most existing in-vehicle AR solutions rely on cameras for head tracking, which suffer from low sampling rate, weak performance at night, and even high computation costs. Wearing a dedicated headset, on the other hand, is intrusive and inconvenient for daily driving.

Wireless signals have been a strong indicator of nearby wireless environment, which can be used in a wide range of applications, including the construction and maintenance of wireless network in smart cities. However, wireless signals can change drastically due to moving pedestrians and automatic power adjustment of Access Points (APs), which renders previous radiomap inaccurate. To achieve sufficient accuracy, surveyors have to update the radiomap constantly, which incurs high maintenance cost in the long run. To address this, we propose RecNet, a neural network to reconstruct a fine-grained radiomap with a small number of new samples. The intuition lies in the visualization of numerical signal strength values by a heatmap. A high-resolution heatmap corresponds to a fine-grained radiomap while a low-resolution one corresponds to a coarse-grained radiomap. Then we reduce the radiomap reconstruction to the image super-resolution: generating a high-resolution image from a low-resolution one. Based on the above, we design and implement the RecNet based on the image super-resolution neural network. Extensive experiments in two large test sites demonstrate that RecNet is able to reconstruct an accurate radiomap with only 50% of fingerprints, and reduces the signal error by more than 20% compared with a recent reconstruction algorithm.

Mobile crowdsensing with increasing pervasiveness of smartphones has enabled a myriad of applications, including urban-scale signal map monitoring and revision. Despite the importance of its quality, due to the large size of a site to cover, dense crowdsourcing is neither cost-effective nor convenient for crowdsourcing participants, making it critical and challenging to balance between signal quality and crowdsourcing cost. To address this problem, we propose a novel incentive mechanism, BCCS, based on Bayesian Compressive Crowdsensing (BCS). BCCS iteratively determines the spatial grids for crowd-sourcing quality and predicts the remaining unexplored grids for deployment efficiency. BCS returns not only the predicted signal values, but also the confidence intervals for convergence and incentive control. A probabilistic user participation and measurement model is applied for incentive design, which is flexible for crowdsensing deployment. Our extensive evaluation based on two different data sets shows that BCCS achieves much higher prediction accuracy (often by more than 20%) with lower payments to the participants and fewer iterations (often by 30%) than existing solutions.

Thanks to their great success as a green urban transportation means of first/last-mile connectivity, bike sharing service (BSS) networks has been proliferating all over the globe. Their station (re)placement and dock resizing has thus become an increasingly important problem for bike sharing service providers.

To improve the accuracy of fingerprint-based localization, one may fuse step counterwith fingerprints. However, the walking step model may vary among people. Such user heterogeneity may lead to measurement error in walking distance. Previous works often require a step counter tediously calibrated offline or through explicit user input. Besides, as device heterogeneity may introduce various signal readings, these studies often need to calibrate the fingerprint RSSI model. Many of them have not addressed how to jointly calibrate the above heterogeneities and locate the user. We propose SLAC, a novel system which simultaneously localizes the user and calibrates the sensors. SLAC works transparently, and is calibration-free with heterogeneous devices and users. Its novel formulation is embedded with sensor calibration, where location estimations, fingerprint signals, and walking motion are jointly optimized with resultant consistent and correct model parameters. To reduce the localization search scope, SLAC first maps the target to a coarse region (say, floor) via stacked denoising autoencoders and then executes the fine-grained localization. Extensive experimental trials at our campus and the international airport further confirm that SLAC accommodates device and user heterogeneity, and outperforms other state-of-the-art fingerprint-based and fusion algorithms by lower localization errors (often by more than 30 percent).

Geomagnetism is promising for indoor localization due to its pervasive presence, high signal stability and unnecessity of extra infrastructure support. It can be fused with a pedometer (step counter) to further improve accuracy. Traditional fusions based on particle filter, however, are computationally complex. Their accuracy is barely satisfactory for large open indoor spaces (such as airports or malls), mainly because particles cannot converge properly in environments with high degree of freedom. They often assume knowledge or explicit user step length input, which is inconvenient or difficult to achieve in reality. We propose Mapel, which efficiently fuses geomagnetism with pedometer based on graphical model without manual user input. Mapel is applicable to any indoor environment. We first discretize indoor map into closely spaced lattice nodes (separated by, say, 0.5~1 meter) in a graph, representing potential target locations. Using conditional random field (CRF), Mapel then conducts geomagnetism-pedometer fusion to estimate user location in the lattice. By jointly considering geomagnetism pattern and step counts, Mapel adaptively self-calibrates a step model for each user without explicit input. Extensive experiments conducted at our campus show that Mapel outperforms state-of-the-art schemes by a wide margin (cutting localization error by more than 40% and reducing computation time by more than 30%).

Site survey cost has become a major deployment challenge for large-scale wireless systems. Some previous approaches require explicit user inputs to label the locations of crowdsourced signals, which is inconvenient in practice. Others may rely on intensively-calibrated specialized sensors. Furthermore, the locations of crowdsourced signals are usually random and sparse, making it difficult to build a complete signal map. To address these, we propose Surecose, a novel signal map construction system by implicit crowdsourcing and interaction of low-cost cooperative sensors. Surecose is initialized with the light-weight sparse signal map collection. Having the background application on, the naive users are implicit surveyors unknowingly uploading crowdsourced signals. Phones may detect each other (mobile sensors), or beacons (fixed sensors). To label their locations, Surecose measures the mutual sensor proximity of the IoT devices. Via joint location labeling and sensor calibration, Surecose transparently locates the users despite devices used. Then Surecose constructs the database with a directionally and environmentally adapted Gaussian process, which scales the labeled locations and RSSIs anywhere into a complete signal map. Extensive experiments in diverse environments, including a campus hall, an international airport and a premium shopping mall, have validated that Surecose can adaptively and accurately construct the Wi-Fi signal map (often more than 20% reduction in RSSI construction error), with low survey cost (around 70% survey reduction) and little human intervention.

Due to its accuracy, trilateration has been widely deployed to locate smartphones outdoors. However, such approach cannot be easily applied indoors due to issues like non-line-of-sight measurement and complex multipath fading. Though fingerprinting overcomes these issues, its accuracy is often hampered by signal noise and the similarity metric comparing signal vectors. We propose INTRI, a novel, simple, accurate, and effective indoor localization framework combining strengths of trilateration and fingerprinting. Given a signal level received from an access point (AP) at target, INTRI first forms a contour given by reference points (RPs) with the same signal level, taking into account signal noise. The target is hence at the juncture of contours formed by all APs. We present selecting RPs for random signal by a width parameter determining the signal contour width (or spread). Then, an LP-based formulation finds the location following spirit of trilateration, which minimizes distance between target position and all contours. A novel particle filter leverages crowdsourced user inputs to adaptively estimate the width parameter. An online algorithm is further used to calibrate heterogeneous smartphones. Our extensive experiments in an airport, a shopping mall, and our campus show INTRI outperforms recent schemes with substantially lower error (often by more than 20 percent).

Wi-Fi fingerprinting has been extensively studied for indoor localization due to its deployability under pervasive indoor WLAN. As the signals from access points (APs) may change due to, for example, AP movement or power adjustment, the traditional approach is to conduct site survey regularly in order to maintain localization accuracy, which is costly and time-consuming. Here, we study how to accurately locate a target and automatically update fingerprints in the presence of altered AP signals (or simply, “altered APs”). We propose Localization with Altered APs and Fingerprint Updating (LAAFU) system, employing implicit crowdsourced signals for fingerprint update and survey reduction. Using novel subset sampling, LAAFU identifies any altered APs and filter them out before a location decision is made, hence maintaining localization accuracy under altered AP signals. With client locations anywhere in the region, fingerprint signals can be adaptively and transparently updated using non-parametric Gaussian process regression. We have conducted extensive experiments in our campus hall, an international airport, and a premium shopping mall. Compared with traditional weighted nearest neighbors and probabilistic algorithms, results show that LAAFU is robust against altered APs, achieving 20 percent localization error reduction with the fingerprints adaptive to environmental signal changes.

In spacious and multi-area buildings, fingerprint-based localization often suffers from expensive location search. Besides, context knowledge like inside/outside-region and floor area is important for complete location service. To address above issues, beyond the algorithms finding the exact location point, we study accurate and efficient indoor area classification for large-scale fingerprint-based system. We first study leveraging the one-class classification to conduct inside/outside-region detection given only the inside fingerprints. Then we discuss different area determination algorithms, and compare their detection accuracy and deployment efficiency. To further enhance accuracy, we also discuss rejecting unclassifiable signals and calibrating heterogeneous devices. We have implemented different algorithms on Android platforms. Experimental trials (totally over 30,000 fingerprints and 15,000 test data) at an international airport, a business building, a premium shopping mall and a university campus have evaluated practicability and deployability of different classification schemes. Our studies can also serve as design guidelines for area classification.

In a wireless system, a signal map shows the signal strength at different locations termed reference points (RPs). As access points (APs) and their transmission power may change over time, keeping an updated signal map is important for applications such as Wi-Fi optimization and indoor localization. Traditionally, the signal map is obtained by a full site survey, which is time-consuming and costly. We address in this paper how to efficiently update a signal map given sparse samples randomly crowdsourced in the space (e.g., by signal monitors, explicit human input, or implicit user participation).

In indoor localization based on Wi-Fi fingerprinting, a target sends its received signal strength indicator (RSSI) of access points (APs) to a server to estimate its position. Traditionally, the server estimates the target position by matching the RSSI with the fingerprints stored in the database. Due to signal noise in fingerprint collection and target measurement, this often results in a geographically disperse set of reference points (RPs), leading to unsatisfactory estimation accuracy. To mitigate the noise problem, we propose a novel, efficient, and highly accurate localization scheme termed Tilejunction. Based on only the first two moments of the measured signal, Tilejunction maps the target RSSI of each AP to a convex hull termed signal “tile” where the target is likely within. Using a novel comparison metric for random signals, we formulate a linear programming (LP) problem to localize the target at the junction of the tiles. To further improve its computational efficiency, Tilejunction employs an information-theoretic measure to keep only those APs whose signals show sufficient differentiation in the site. It also partitions the site into multiple clusters to substantially reduce the search space in the LP optimization. We have implemented Tilejunction. Our extensive simulation and experimental measurements show that it outperforms other recent state-of-the-art approaches (e.g. RADAR, KL-divergence, etc.) with significantly lower localization error (often by more than 30 percent).

In order to improve the accuracy of fingerprint-based localization, one may fuse step counter measurement with location estimation. Previous works on this often require a pre-calibrating the step counter with training sequence or explicit user input, which is inconvenient for practical deployment. Some assume conditional independence on successive sensor readings, which achieves unsatisfactory accuracy in complex and noisy environment. Some other works need a calibration process for RSSI measurement consistency if different devices are used for offline fingerprint collection and online location query.

Fusing fingerprints with mutual distance information potentially improves indoor localization accuracy. Such distance information may be spatial (e.g., via inter-node measurement) or temporal (e.g., via dead reckoning). Previous approaches on distance fusion often require exact distance measurement, assume the knowledge of distance distribution, or apply narrowly to some specific sensing technology or scenario. Due to random signal fluctuation, wireless fingerprints are inherently noisy and distance cannot be exactly measured. We hence propose Wi-Dist, a highly accurate indoor localization framework fusing noisy fingerprints with uncertain mutual distances (given by their bounds). Wi-Dist is a generic framework applicable to a wide range of sensors (peer-assisted, INS, etc.) and wireless fingerprints (Wi-Fi, RFID, CSI, etc.). It achieves low errors by a convex-optimization formulation which jointly considers distance bounds and only the first two moments of measured fingerprint signals. We implement Wi-Dist, and conduct extensive simulation and experimental studies based on Wi-Fi in our international airport and university campus. Our results show that Wi-Dist achieves significantly better accuracy than other state-of-the-art schemes (often by more than 40%).

Trilateration has been widely and successfully employed to locate outdoor mobile devices due to its accuracy. However, it cannot be directly applied for indoor localization due to issues such as non-line-of-sight measurement and multipath fading. Though fingerprinting overcomes these issues, its accuracy is often hampered by signal noise and the choice of similarity metric between signal vectors. We propose INTRI, a novel, simple and effective indoor localization technique combining the strengths of trilateration and fingerprinting.

In Wi-Fi fingerprint localization, a target sends its measured Received Signal Strength Indicator (RSSI) of access points (APs) to a server for its position estimation. Traditionally, the server estimates the target position by matching the RSSI with the fingerprints stored in database. Due to signal measurement uncertainty, this matching process often leads to a geographically dispersed set of reference points, resulting in unsatisfactory estimation accuracy. We propose a novel, efficient and highly accurate localization scheme termed Sectjunction which does not lead to a dispersed set of neighbors. For each selected AP, Sectjunction sectorizes its coverage area according to discrete signal levels, hence achieving robustness against measurement uncertainty. Based on the received AP RSSI, the target can then be mapped to the sector where it is likely to be. To further enhance its computational efficiency, Sectjunction partitions the site into multiple area clusters to narrow the search space. Through convex optimization, the target is localized based on the cluster and the junction of the sectors it is within. We have implemented Sectjunction, and our extensive experiments show that it significantly outperforms recent schemes with much lower estimation error.