Leveraging cross-view geo-localization with ensemble learning and temporal awareness.

The Global Navigation Satellite System (GNSS) is unreliable in some situations. To mend the poor GNSS signal, an autonomous vehicle can self-localize by matching a ground image against a database of geotagged aerial images. However, this approach has challenges because of the dramatic differences in the viewpoint between aerial and ground views, harsh weather and lighting conditions, and the lack of orientation information in training and deployment environments. In this paper, it is shown that previous models in this area are complementary, not competitive, and that each model solves a different aspect of the problem. There was a need for a holistic approach. An ensemble model is proposed to aggregate the predictions of multiple independently trained state-of-the-art models. Previous state-of-the-art (SOTA) temporal-aware models used heavy-weight network to fuse the temporal information into the query process. The effect of making the query process temporal-aware is explored and exploited by an efficient meta block: naive history. But none of the existing benchmark datasets was suitable for extensive temporal awareness experiments, a new derivative dataset based on the BDD100K dataset is generated. The proposed ensemble model achieves a recall accuracy R@1 (Recall@1: the top most prediction) of 97.74% on the CVUSA dataset and 91.43% on the CVACT dataset (surpassing the current SOTA). The temporal awareness algorithm converges to R@1 of 100% by looking at a few steps back in the trip history.

Driving behavior classification at signalized intersections using vehicle kinematics: Application of unsupervised machine learning.

Driving behavior is considered as a unique driving habit of each driver and has a significant impact on road safety. This study proposed a novel data-driven Machine Learning framework that can classify driving behavior at signalized intersections considering two different signal conditions. To the best of our knowledge, this is the first study that investigates driving behavior at signalized intersections with two different conditions that are mostly used in practice, i.e., the control setting with the signal order of green-yellow-red and a flashing green setting with the signal order of green-flashing green-yellow-red. A driving simulator dataset collected from participants at Qatar University's Qatar Transportation and Traffic Safety Center, driving through multiple signalized intersections, was used. The proposed framework extracts volatility measures from vehicle kinematic parameters including longitudinal speed and acceleration. K-means clustering algorithm with elbow method was used as an unsupervised machine learning to cluster driving behavior into three classes (i.e., conservative, normal, and aggressive) and investigate the impact of signal conditions. The framework confirmed that in general driving behavior at a signalized intersection reflects drivers' habits and personality rather than the signal condition, still, it manifests the intersection nature that usually requires drivers to be more vigilant and cautious. Nonetheless, the results suggested that flashing green condition could make drivers more conservative, which could be due to the limited capabilities of human to estimate the remaining distance and the prolonged duration of the additional flashing green interval. The proposed framework and findings of the study were promising that can be used for clustering drivers into different styles for different conditions and might be beneficial for policymakers, researchers, and engineers.

The COVID-19 impacts on bikeshare systems in small rural communities: Case study of bikeshare riders in Montgomery County, VA.

The shared and micro-mobility industry (ride sharing and hailing, carpooling, bike and e-scooter shares) saw direct and almost immediate impacts from COVID-19 restrictions, orders and recommendations from local governments and authorities. However, the severity of that impact differed greatly depending on variables such as different government guidelines, operating policies, system resiliency, geography and user profiles. This study investigated the impacts of the pandemic regarding bike-share travel behavior in Montgomery County, VA. We used bike-usage dataset covering two small towns in Montgomery county, namely: Blacksburg and Christiansburg, including Virginia Tech campus. The dataset used covers the period of Jan 2019-Dec 2021 with more than 14,555 trips and 5,154 active users. Findings indicated that a bikeshare user's average trip distance and duration increased in 2020 (compared to 2019) from 2+ miles to 4+ and from half an hour to about an hour. While there was a slight drop in 2021, bikeshare users continued to travel farther distances and spend more time on the bikes than pre-COVID trips. When those averages were unpacked to compare weekday trips to weekend trips, a few interesting trip patterns were observed. Unsurprisingly, more trips still took place on the weekends (increasing from 2x as many trips to 4x as many trips than the weekday). These findings could help to better understand traveler's choices and behavior when encountering future pandemics.

Solving pickup and drop-off problem using hybrid pointer networks with deep reinforcement learning.

In this study, we propose a general method for tackling the Pickup and Drop-off Problem (PDP) using Hybrid Pointer Networks (HPNs) and Deep Reinforcement Learning (DRL). Our aim is to reduce the overall tour length traveled by an agent while remaining within the truck's capacity restrictions and adhering to the node-to-node relationship. In such instances, the agent does not allow any drop-off points to be serviced if the truck is empty; conversely, if the vehicle is full, the agent does not allow any products to be picked up from pickup points. In our approach, this challenge is solved using machine learning-based models. Using HPNs as our primary model allows us to gain insight and tackle more complicated node interactions, which simplified our objective to obtaining state-of-art outcomes. Our experimental results demonstrate the effectiveness of the proposed neural network, as we achieve the state-of-art results for this problem as compared with the existing models. We deal with two types of demand patterns in a single type commodity problem. In the first pattern, all demands are assumed to sum up to zero (i.e., we have an equal number of backup and drop-off items). In the second pattern, we have an unequal number of backup and drop-off items, which is close to practical application, such as bike sharing system rebalancing. Our data, models, and code are publicly available at Solving Pickup and Dropoff Problem Using Hybrid Pointer Networks with Deep Reinforcement Learning.

Hybrid pointer networks for traveling salesman problems optimization.

In this work, we proposed a hybrid pointer network (HPN), an end-to-end deep reinforcement learning architecture is provided to tackle the travelling salesman problem (TSP). HPN builds upon graph pointer networks, an extension of pointer networks with an additional graph embedding layer. HPN combines the graph embedding layer with the transformer's encoder to produce multiple embeddings for the feature context. We conducted extensive experimental work to compare HPN and Graph pointer network (GPN). For the sack of fairness, we used the same setting as proposed in GPN paper. The experimental results show that our network significantly outperforms the original graph pointer network for small and large-scale problems. For example, it reduced the cost for travelling salesman problems with 50 cities/nodes (TSP50) from 5.959 to 5.706 without utilizing 2opt. Moreover, we solved benchmark instances of variable sizes using HPN and GPN. The cost of the solutions and the testing times are compared using Linear mixed effect models. We found that our model yields statistically significant better solutions in terms of the total trip cost. We make our data, models, and code publicly available https://github.com/AhmedStohy/Hybrid-Pointer-Networks.

Crash severity analysis of vulnerable road users using machine learning.

Road crash fatality is a universal problem of the transportation system. A massive death toll caused annually due to road crash incidents, and among them, vulnerable road users (VRU) are endangered with high crash severity. This paper focuses on employing machine learning-based classification approaches for modelling injury severity of vulnerable road users-pedestrian, bicyclist, and motorcyclist. Specifically, this study aims to analyse critical features associated with different VRU groups-for pedestrian, bicyclist, motorcyclist and all VRU groups together. The critical factor of crash severity outcomes for these VRU groups is estimated in identifying the similarities and differences across different important features associated with different VRU groups. The crash data for the study is sourced from the state of Queensland in Australia for the years 2013 through 2019. The supervised machine learning algorithms considered for the empirical analysis includes the K-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Forest (RF). In these models, 17 distinct road crash parameters are considered as input features to train models, which originate from road user characteristics, weather and environment, vehicle and driver condition, period, road characteristics and regions, traffic, and speed jurisdiction. These classification models are separately trained and tested for individual and unified VRU to assess crash severity levels. Afterwards, model performances are compared with each other to justify the best classifier where Random Forest classification models for all VRU modes are found to be comparatively robust in test accuracy: (motorcyclist: 72.30%, bicyclist: 64.45%, pedestrian: 67.23%, unified VRU: 68.57%). Based on the Random Forest model, the road crash features are ranked and compared according to their impact on crash severity classification. Furthermore, a model-based partial dependency of each road crash parameters on the severity levels is plotted and compared for each individual and unified VRU. This clarifies the tendency of road crash parameters to vary with different VRU crash severity. Based on the outcome of the comparative analysis, motorcyclists are found to be more likely exposed to higher crash severity, followed by pedestrians and bicyclists.

A Novel Crowdsourcing Model for Micro-Mobility Ride-Sharing Systems.

Substantial research is required to ensure that micro-mobility ride sharing provides a better fulfilment of user needs. This study proposes a novel crowdsourcing model for the ride-sharing system where light vehicles such as scooters and bikes are crowdsourced. The proposed model is expected to solve the problem of charging and maintaining a large number of light vehicles where these efforts will be the responsibility of the crowd of suppliers. The proposed model consists of three entities: suppliers, customers, and a management party responsible for receiving, renting, booking, and demand matching with offered resources. It can allow suppliers to define the location of their private e-scooters/e-bikes and the period of time they are available for rent. Using a dataset of over 9 million e-scooter trips in Austin, Texas, we ran an agent-based simulation six times using three maximum battery ranges (i.e., 35, 45, and 60 km) and different numbers of e-scooters (e.g., 50 and 100) at each origin. Computational results show that the proposed model is promising and might be advantageous to shift the charging and maintenance efforts to a crowd of suppliers.

A review of naturalistic driving study surrogates and surrogate indicator viability within the context of different road geometries.

Advancements in data collection and processing methods have produced large databases containing high quality vehicular data. Despite this, conventional vehicle-vehicle collisions remain difficult to identify due to their rarity. Therefore, there is a need to identify potential collisions given the introduction of these new data collection methods. Surrogate indicators are a popular methods utilised to identify such events, however, the type of surrogate that can be used depends heavily on the type of data collection method. Though most surrogate indicators are used at different road geometries, there is evidence to suggest that some surrogate indicators may perform better than others at a given geometry. This review provides two key contributions to the body of literature. Firstly, a review of kinematic surrogates is put forward, along with a discussion on the whether these surrogates can be contextualised at different road geometries. Secondly, an extensive analysis and discussion of observer-based and video processed surrogate indicators, the collision types they aim to identify and the geometries they have been used at previously were analysed and advantages and disadvantages of the surrogates have been presented for future use. To do this, intersections, highways and roundabouts were selected and divided into geometry subtypes (i.e. three-legged and four-legged intersection) and segments (i.e. approaches to intersections and internal to the intersection) based on the likelihood of crash types and pre-crash manoeuvres occurring in that segment. Due to the lack of research around the use of kinematic triggers at road geometries, it is difficult to advocate for the use of any given trigger over another at a given geometry. Furthermore, it was found that kinematic triggers cannot accurately identify conflicts from naturalistic driving data and require the use of advanced statistical techniques such as machine learning to increase accuracy. A brief analysis of threshold identification techniques was also performed. Several future works have been put forward including the introduction of surrogates which capture conflict severity and the role of surrogate indicators in connected and automated vehicle environments.

Correction: A framework for testing independence between lane change and cooperative intelligent transportation system.

[This corrects the article DOI: 10.1371/journal.pone.0229289.].

A framework for testing independence between lane change and cooperative intelligent transportation system.

Cooperative Intelligent Transportation Systems (C-ITS) are being deployed in several cities around the world. We are preparing for the largest Field Operational Test (FOT) in Australia to evaluate C-ITS safety benefits. Two of the safety benefit hypotheses we formulated assume a dependency between lane changes and C-ITS warnings displayed on the Human Machine Interface (HMI) during safety events. Lane change detection is done by processing many predictors from several sensors at the time of the safety event. However, in our planned FOT, the participating vehicles are only equipped with the vehicle C-ITS and the IMU. Therefore, in this paper, we propose a framework to test lane change and C-ITS dependency. In this framework, we train a random forest classifier using data collected from the IMU to detect lane changes. Consequently, the random forest output probabilities of the testing data in case of C-ITS and control are used to construct a 2x2 contingency table. Then we develop a permutation test to calculate the null hypothesis needed to test the independence of the lane change during safety events and the C-ITS.

Modeling driver stop/run behavior at the onset of a yellow indication considering driver run tendency and roadway surface conditions.

The ability to model driver stop/run behavior at signalized intersections considering the roadway surface condition is critical in the design of advanced driver assistance systems. Such systems can reduce intersection crashes and fatalities by predicting driver stop/run behavior. The research presented in this paper uses data collected from two controlled field experiments on the Smart Road at the Virginia Tech Transportation Institute (VTTI) to model driver stop/run behavior at the onset of a yellow indication for different roadway surface conditions. The paper offers two contributions. First, it introduces a new predictor related to driver aggressiveness and demonstrates that this measure enhances the modeling of driver stop/run behavior. Second, it applies well-known artificial intelligence techniques including: adaptive boosting (AdaBoost), random forest, and support vector machine (SVM) algorithms as well as traditional logistic regression techniques on the data in order to develop a model that can be used by traffic signal controllers to predict driver stop/run decisions in a connected vehicle environment. The research demonstrates that by adding the proposed driver aggressiveness predictor to the model, there is a statistically significant increase in the model accuracy. Moreover the false alarm rate is significantly reduced but this reduction is not statistically significant. The study demonstrates that, for the subject data, the SVM machine learning algorithm performs the best in terms of optimum classification accuracy and false positive rates. However, the SVM model produces the best performance in terms of the classification accuracy only.