Subject Vehicle Research Articles

Mechanism and modeling of car-following behavior under local multi-vehicle influence

Aiming at the limitations of single-dimensional car-following model to describe vehicle car-following behavior in local multi-vehicle environment, the mechanism of vehicles in adjacent lane influencing the subject vehicle car-following behavior is explored, and a vehicle car-following model more suitable for local multi-vehicle environment is attempted to be established. The driving behavior variable reflecting the influence of vehicles in adjacent lane was determined by correlation analysis. Vortisch indicator of similarity(VIS) was used to characterize the influence of vehicles in adjacent lane on car-following behavior. Chi-square independent test and kernel density curve were used to determine the VIS demarcation threshold reflecting whether the influence is significant. Recursive feature elimination was used to screen the variables related to car-following samples significantly affected by vehicles in adjacent lane. The influence mechanism of variables was determined according to the statistical analysis results. Based on the mechanism proposed, the car-following model suitable for local multi-vehicle environment was constructed and its prediction effect was evaluated. Results show that VIS between the speed of subject vehicle and following vehicle in adjacent lane can characterize the influence and the VIS threshold is 0.668. It is concluded that the attention mechanism and memory effect can explain the influence mechanism of car-following behavior in multi-vehicle environment. Considering the attention mechanism and memory effect, RMSE decrease by 65%in Full Velocity Difference Model (FVD) model and 62%in Intelligent Driver Model (IDM) model, MAE decrease by 65%and 59%and R2 increased by 180%and 288%respectively, which proved the rationality of the attention mechanism and memory effect in explaining the car following behavior of subject vehicle in local multi-vehicle environment.

Driving Characteristics of Drivers Avoiding Lateral Obstacles in Mountain Highway Tunnels: A Dynamic Lateral Safety Distance Prediction Model

A natural driving real-vehicle test was conducted in nine tunnels in Chongqing, China. The driving trajectories and speeds of 40 drivers were recorded to better understand drivers’ driving characteristics under the mutual influence of various tunnel lengths and numerous lateral obstacles on mountainous highways. According to the driver’s need for lateral safety distance in different scenarios, the conditions and scope of the occurrence of the shy away effect are defined, and a dynamic avoidance prediction model is established in multiple scenarios. It was demonstrated that, when there were no lateral dynamic obstacles, drivers were influenced by the tunnel sidewalls to drift away from them. However, when lateral dynamic obstacles were present, drivers’ driving trajectories were corrected to be closer to the inside of the lane. The influence of vehicles of the same type resulted in the trajectory offset to the left by 0.2 m (extra-long tunnel), 0.7 m (long tunnel), 0.5 m (medium tunnel), and 0.7 m (short tunnel) compared with the driving trajectory without the influence of the obstacle. The influence of trucks causes the trajectory offset to the left compared with the driving trajectory without obstacles by 0.2 m (extra-long tunnel), 0.5 m (long tunnel), 1 m (medium tunnel), and 1.75 m (short tunnel). There is some variation in the degree of risk to the driver from different obstacles under different tunnels. The lateral distance between the driver and the dynamic obstacle is related to the nature of the available driving distance ahead. A sufficiently long driving maneuver distance enables the subject vehicle to gradually overtake the lateral dynamic obstacle in a more stable state. Dynamic avoidance models with fitting accuracies of 0.6–0.99 were implemented to predict vehicle trajectories in different tunnels and under the influence of various lateral dynamic obstacles. The study of the correlation of vehicle trajectories under the influence of various factors in tunnels is completed by the research findings, which can serve as a foundation for the design of driving behavior regulation, an improvement in traffic safety facilities, traffic management, and monitoring of lateral safety distances of intelligent vehicles in tunnels.

Unveiling gap acceptance behaviour during lane change with EDIV data: A deep dive into driving behaviour on expressway using a three level mixed effect linear regression approach

Lane change has a potential significance in road safety. Gap acceptance phenomena serves as a primary and critical phase in lane change maneuver. This study aims to investigate the gap acceptance behaviour of drivers during lane changes on expressways, with a focus on understanding how various factors influence drivers' decisions to change lanes. An extensive dataset collected through various sensors tailored for expressway driving, known as the ‘Expressway Drive: Instrumented Vehicle (EDIV) Dataset’ is utilized. Driving data from 59 drivers covering a distance of around 4000 km was used in the current study. Total 2578 lane changing events are identified through computing lateral deviations measured through 3D LiDAR sensor. Substantial differences are observed within the groups in primary analysis which suggest that lane-change direction significantly affect gap acceptance. To effectively manage both intra- and inter-cluster variances, this study employs two separate three levels mixed-effects linear models. These models account for the interdependence of gap acceptance characteristics within individual drivers and for different directions of lane changes by incorporating random effects. Furthermore, these models examine relationships between lead/ lag gap acceptance and the various influencing factors as fixed effects. It was found that factors such as speed of the subject vehicle, gap position, relative speeds, and surrounding vehicle types had influence on gap acceptance during lane changes on expressways. The insights gained from this study could inform the development of advanced driver assistance systems (ADAS) as well as development of autonomous vehicles, contributing to improved road safety and traffic flow management in high-speed environments.

Fuzzy inference systems for discretionary lane changing decisions: Model improvements and research challenges

Lane Changing Decision Model (LCDM) is a critical component in semi- and fully-automated driving systems. Recent research has found that Fuzzy Inference System (FIS) is a promising approach to implementing LCDMs. To improve the FIS’s performance, this research reviewed the challenges in the development of a FIS model to make the yes,no decisions in discretionary lane changes. The FIS model was revised to bring its fuzzy inference rules more consistent with the fuzzy membership functions, and its composition and defuzzification methods more in line with the classical fuzzy logic theory. An equitable test data set with approximately equal number of yes,no data points was assembled from the same NGSIM data used in the past research. The test results proved that: (1) A LCDM’s performance was dependent on how the yes,no decisions in the test data set were manually labeled; and (2) Separating the fuzzy inference rules into a yes group and a no group and compute the results separately yielded potentially better decision accuracy. Furthermore, The Genetic Expression Programming Model (GEPM) performed better than the improved FIS-based model. The findings led the authors to suggest two possible research directions: (1) Add the subject vehicle’s speed as an input to the LCDM and redesign the decision-making model; and (2) Construct models for congested and uncongested traffic separately. The authors further suggested the use of instrumented vehicles to collect a set of high-fidelity lane changing data in the naturalistic driving environment. (word count: 246)

A methodology of cooperative driving based on microscopic traffic prediction

We present a methodology of cooperative driving in vehicular traffic, in which for short-time traffic prediction rather than one of the statistical approaches of artificial intelligence (AI), we follow a qualitative different microscopic traffic prediction approach developed recently (Kerner and Klenov, 2022). In the microscopic traffic prediction approach used for the planning of the subject vehicle trajectory, no learning algorithms of AI are applied; instead, microscopic traffic modeling based on the physics of vehicle motion is used. The presented methodology of cooperative driving is devoted to application cases in which microscopic traffic prediction without cooperative driving cannot lead to a successful vehicle control and trajectory planning. For the understanding of the physical features of the methodology of cooperative driving, a traffic city scenario has been numerically studied, in which a subject vehicle, which requires cooperative driving, is an automated vehicle. Based on microscopic traffic prediction, in the methodology first a cooperating vehicle(s) is found; then, motion requirements for the cooperating vehicle(s) and characteristics of automated vehicle control are predicted and used for vehicle motion; to update predicted characteristics of vehicle motion, calculations of the predictions of motion requirements for the cooperating vehicle and automated vehicle control are repeated for each next time instant at which new measured data for current microscopic traffic situation are available. With the use of microscopic traffic simulations, the evaluation of the applicability of this methodology is illustrated for a simple case of unsignalized city intersection, when the automated vehicle wants to turn right from a secondary road onto the priority road.

Deep Reinforcement Learning Lane-Changing Decision Algorithm for Intelligent Vehicles Combining LSTM Trajectory Prediction

Intelligent decisions for autonomous lane-changing in vehicles have consistently been a focal point of research in the industry. Traditional lane-changing algorithms, which rely on predefined rules, are ill-suited for the complexities and variabilities of real-world road conditions. In this study, we propose an algorithm that leverages the deep deterministic policy gradient (DDPG) reinforcement learning, integrated with a long short-term memory (LSTM) trajectory prediction model, termed as LSTM-DDPG. In the proposed LSTM-DDPG model, the LSTM state module transforms the observed values from the observation module into a state representation, which then serves as a direct input to the DDPG actor network. Meanwhile, the LSTM prediction module translates the historical trajectory coordinates of nearby vehicles into a word-embedding vector via a fully connected layer, thus providing predicted trajectory information for surrounding vehicles. This integrated LSTM approach considers the potential influence of nearby vehicles on the lane-changing decisions of the subject vehicle. Furthermore, our study emphasizes the safety, efficiency, and comfort of the lane-changing process. Accordingly, we designed a reward and penalty function for the LSTM-DDPG algorithm and determined the optimal network structure parameters. The algorithm was then tested on a simulation platform built with MATLAB/Simulink. Our findings indicate that the LSTM-DDPG model offers a more realistic representation of traffic scenarios involving vehicle interactions. When compared to the traditional DDPG algorithm, the LSTM-DDPG achieved a 7.4% increase in average single-step rewards after normalization, underscoring its superior performance in enhancing lane-changing safety and efficiency. This research provides new ideas for advanced lane-changing decisions in autonomous vehicles.

Deep Reinforcement Learning Car-Following Control Based on Multivehicle Motion Prediction

Reinforcement learning (RL)–based car-following (CF) control strategies have attracted significant attention in academia, emerging as a prominent research topic in recent years. Most of these control strategies focus solely on the motion status of the immediately preceding vehicle. However, with the development of vehicle-to-vehicle (V2V) communication technologies, intelligent vehicles such as connected autonomous vehicles (CAVs) can gather information about surrounding vehicles. Therefore, this study proposes an RL-based CF control strategy that takes multivehicle scenarios into account. First, the trajectories of two preceding vehicles and one following vehicle relative to the subject vehicle (SV) are extracted from a highD dataset to construct the environment. Then the twin-delayed deep deterministic policy gradient (TD3) algorithm is implemented as the control strategy for the agent. Furthermore, a sequence-to-sequence (seq2seq) module is developed to predict the uncertain motion statuses of surrounding vehicles. Once integrated into the RL framework, this module enables the agent to account for dynamic changes in the traffic environment, enhancing its robustness. Finally, the performance of the CF control strategy is validated both in the highD dataset and in two traffic perturbation scenarios. In the highD dataset, the TD3-based prediction CF control strategy outperforms standard RL algorithms in terms of convergence speed and rewards. Its performance also surpasses that of human drivers in safety, efficiency, comfort, and fuel consumption. In traffic perturbation scenarios, the performance of the proposed CF control strategy is compared with the model predictive controller (MPC). The results show that the TD3-based prediction CF control strategy effectively mitigates undesired traffic waves caused by the perturbations from the head vehicle. Simultaneously, it maintains the desired traffic state and consistently ensures a stable and efficient traffic flow.

Application of machine learning models to predict driver left turn destination lane choice behavior at urban intersections

When there are multiple lanes to choose from downstream of a turning movement, drivers should choose the innermost lane so that drivers at other approaches of the intersection may make concurrent turning movements in the outermost lane(s). However, human drivers do not always choose the innermost lane, which could lead to crashes with other vehicles. Therefore, predicting human driver behaviors is vital in reducing crashes, as the need to share the roadways with automated vehicles (AVs) continues to grow. In this research, various machine learning models have been used to predict the left turn destination lane choice of human-driven vehicles (HDVs) at urban intersections based on several quantifiable parameters. A total of 174 subject vehicles were extracted and analyzed in Los Angeles, California, and Atlanta, Georgia, using HDV trajectory data from the Next Generation SIMulation (NGSIM) database. Five machine learning techniques, namely binary logistic regression, k nearest neighbors, support vector machines, random forest, and adaptive neuro-fuzzy inference system, were applied to the extracted data to predict the lane choice behavior of drivers. The k nearest neighbors model showed the most promising results for the evaluated data with a correct decision score of over 93% for the unseen test data. This model may be programmed into: (i) AVs, in conjunction with sensors, to predict if an HDV is about to turn into the incorrect destination lane; and (ii) microscopic traffic simulation tools so that modelers can identify potential conflicts when HDVs do not select the appropriate destination lane.

Preparing drivers for the future: Evaluating the effects of training on drivers’ performance in an autonomous vehicle landscape

The rapid development of Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicle (AV) technology has raised concerns about the ability of ordinary drivers to safely interact with these systems without formal training or practice sessions. This study investigates the impact of training on drivers' ability to interact with ADAS functions, aiming to enhance road safety and prepare drivers for future autonomous driving. To do so the study compared novice and experienced drivers' performance across various driving training sessions separated by different time intervals. Using a simulation-based driving scenario, a diverse group of participants with different ages, genders, driving experience, and ADAS usage frequencies was recruited. Participants were instructed to drive a virtual subject vehicle (SV) multiple times with different time intervals and various ADAS functions being activated and/or deactivated. Data on steering wheel input, acceleration, braking, event triggering time, response time, and vehicle lateral position were collected and analysed. Participants' performance, measured in terms of accuracy, reaction time in interacting with ADAS functions, and vehicle control when they are interacting with those functions, was assessed after each training session. Results revealed significant improvements in accuracy, reaction time, and vehicle control for both novice and experienced drivers' performance following training. The results highlight the importance of appropriate training for drivers to safely operate ADAS and autonomous functions, providing valuable insights for policymakers, vehicle manufacturers, and driver education institutions. Furthermore, the findings emphasise the need for further investigations on the effectiveness of various training modalities and the potential benefits of integrating real-time feedback systems or Artificial Intelligence (AI)-assisted coaching. This study has important ramifications for the future design and implementation of driver training programs, as well as the development of policies and regulations to promote the safe use of ADAS and AV technology in the future.

A physics-informed Transformer model for vehicle trajectory prediction on highways

Autonomous Vehicles (AVs) have made remarkable developments and are anticipated to replace human drivers. In transitioning from human-driven vehicles to fully AVs, one crucial task is to predict the trajectories of the subject vehicle and its surrounding vehicles in real time. Most existing methods of vehicle trajectory prediction on highways are based on physical models or purely data-driven models. However, they either yield unsatisfactory prediction performance or lack model interpretability and physical implications. This paper proposes a Physics-Informed Deep Learning framework that fully leverages the advantages of data-driven and physics-based models to go beyond the existing models. We use the Transformer neural network architecture with self-attention as Physics-Uninformed Neural Network (PUNN) and Intelligent Driver Model (IDM) as physical model to construct of Physics-Informed Transformer-Intelligent Driver Model (PIT-IDM). Extensive experiments have been conducted on two datasets with different traffic environments, i.e., Next Generation SIMulation (NGSIM) data in the US and the Ubiquitous Traffic Eyes (UTE) data in China, to verify model accuracy and efficiency. Compared with the three kinds of baselines by relative and absolute measures of effectiveness, the best performing PIT-IDM reduces longitudinal trajectory prediction errors for long horizons by 5%-50%, some even reduced up to 70%. Extensive empirical analyses have been carried out to verify its excellent spatio-temporal transferability and explore the physics-informed mechanism underlying this deep learning method. The training and inference time analysis indicates that although it takes longer to train PIT-IDM, it requires fewer calls and accumulates fewer errors with less computation time in real-world applications. The overall results further validate the efficacy of this Physics-Informed Deep Learning framework in enhancing model accuracy, interpretability, and transferability.

Energy management of plug-in hybrid electric vehicles based on speed prediction fused driving intention and LIDAR

Driving intention and speed prediction are essential factors in the energy management of plug-in hybrid electric vehicles (PHEVs). This paper proposes an improved energy management strategy for the subject vehicle by speed prediction fused with driving intention and LIDAR data in a vehicle-following scenario. A driving intention recognition model is developed based on the gated recurrent unit (GRU), which takes the vehicle speed, throttle opening, and brake pedal force of the subject vehicle as input. Then integrating the LIDAR point cloud data and driving intention result of the subject vehicle to achieve more accurate speed prediction, where joint probabilistic data association and interacting multiple models methods are used to process LIDAR data. The more accurate speed prediction is then applied to design a prediction-informed adaptive equivalent consumption minimization strategy (PIA-ECMS) for real-time energy management optimization. Experimental results demonstrate the recognition accuracy of up to 88%, indicating that the driver’s driving intention can be identified effectively. The speed prediction has an error margin of no more than 5.9 km/h. Compared with existing adaptive ECMS without speed prediction, the proposed PIA-ECMS can enhance fuel economy by 1.3–2.7% while achieving better SOC charge sustainability.

Evaluation of intersection crashes using naturalistic driving data through the lens of future I-ADAS

Objective Intersection advanced driver assistance systems (I-ADAS) with the capability to detect possible collisions and perform evasive braking have the potential to reduce the number of intersection crashes. However, these systems will encounter many challenges caused by the complexity of real-world driving conditions. The purpose of this study is to use real-world naturalistic driving data to conduct an initial exploration of the potential challenges for future I-ADAS in straight crossing path (SCP), left turn across path/lateral direction (LTAP/LD), and left turn across path/opposite direction (LTAP/OD) crash configurations. Methods Intersection crashes were selected from the Second Strategic Highway Research Program (SHRP 2) Naturalistic Driving Study. The SHRP 2 dataset includes front-facing, driver-facing, rear-facing, and a hands/feet-facing video and vehicle speed, steering, accelerator, and brake time-series data. This data was reviewed to understand driver sightline obstructions, driver distractions, and initiation of driver responses. The estimated time to collision (TTC) from the precipitating event, defined as when either vehicle entered the intersection without the right-of-way, was computed based on the distance to the impact point divided by the current velocity of the subject vehicle. Results The median impact speed was 18.0 km/h for SCP and LTAP/LD crashes and 16.1 km/h for LTAP/OD crashes. The median TTC from the precipitating event was 1.35 s for SCP and LTAP/LD crashes and 1.44 s for LTAP/OD crashes. For SCP crashes, the three main sightline obstruction scenarios were slower vehicles traveling in the same direction waiting to turn right, vehicles in the closer crossing lane, and a parked truck. For LTAP/OD crashes, the sightline obstruction was often oncoming vehicles in a closer lane blocking the view of another vehicle. Conclusion Sightline obstructions could present a challenge for future I-ADAS to activate in SCP, LTAP/LD, and LTAP/OD crashes. This study utilized naturalistic driving data to complete a comprehensive analysis of intersection crashes, including driver distractions, evasive maneuvers, and sightline obstructions that can assist in the development of I-ADAS. This analysis is not possible with police-reported crash data only, which does not contain necessary details on the driver and surrounding environment.

Drivers’ Dilemma at High-Speed Unsignalized Intersections

At unsignalized intersections, drivers typically reject smaller gaps and accept larger gaps in traffic. However, drivers experience a dilemma or confusion over a wide range of gaps, since incorrect decisions by them can lead to crashes. The present study quantifies the drivers’ dilemma at high-speed unsignalized intersections. Traffic data on gap size (temporal and spatial), driver’s decision (acceptance or rejection), the waiting time of the offending vehicle, the offending and conflicting vehicle types, and the speed of the conflicting vehicle are extracted from recorded video. The decisions by drivers to accept or reject gaps are modeled as a function of the gap size, the waiting time of the offending vehicle, and the speed of the conflicting vehicle using binary logit regression. The results reveal that the probability of rejection decreases as the gap size and waiting time of the subject vehicle increases. On the contrary, as the speed of the conflicting vehicle increases, the probability of rejection increases. The length and location of the dilemma zone were investigated using vehicle type, right-turning movement, and the speed of the conflicting vehicle, and a significant effect was noted. Moreover, the length and location of the dilemma zone are analytically quantified based on drivers’ compliance with a driver assistance system (DAS). The length of the dilemma decreases as the proportion of drivers’ compliance with the DAS increases. Further, the location of the dilemma zone shifts upstream of the intersection with an increase in drivers’ compliance with the DAS.

A new lane change index for lane change conflicts at weaving segments

Objective Surrogate safety measures provide information about the safety of traffic facilities without waiting for crashes to occur. Most studies related to surrogate safety measures were performed at intersections. Traffic accident statistics show that sideswipe crashes have a considerable percentage of all crashes. Sideswipe crashes arising from lane-change conflicts are highly probable at weaving segments. This study focuses on lane-change conflicts at weaving segments and proposes a new index. Methods Considering the interactions between the subject vehicle and the surrounding vehicles, two new surrogate safety measures, time- and position-based, are recommended. Then, applying a fault tree analysis, these two measures are converted to a lane change risk index to estimate the crash severity. The proposed index is specific to weaving segments. Data from five weaving segments were extracted using deep learning detection and tracking algorithms. Results The results based on the relationship between the proposed indexes and speed showed that the probability and severity of a crash increased as the speed increased. The longer weaving length resulted in smaller values in the time-based index. Conclusions The weaving length is critical because it becomes difficult to perform a lane change if the length is too short. The proposed time and position-based indexes combine the effect of weaving length, stopping distance index, time to collision, and brake time. It is believed that the proposed lane change risk index can be used to detect the safety level of weaving segments.

A Finite-Sampling, Operational Domain Specific, and Provably Unbiased Connected and Automated Vehicle Safety Metric

A connected and automated vehicle safety metric determines the performance of a subject vehicle (SV) by analyzing the data involving the interactions among the SV and other dynamic road users and environmental features. When the data set contains only a finite set of samples collected from the naturalistic mixed-traffic driving environment, a metric is expected to generalize the safety assessment outcome from the observed finite samples to the unobserved cases by specifying in what domain the SV is expected to be safe and how safe the SV is, statistically, in that domain. However, to the best of our knowledge, none of the existing safety metrics are able to justify the above properties with an operational domain specific, guaranteed complete, and provably unbiased safety evaluation outcome. In this paper, we propose a novel safety metric that involves the $\alpha$-shape and the $\epsilon$-almost robustly forward invariant set to characterize the SV's almost safe operable domain and the probability for the SV to remain inside the safe domain indefinitely, respectively. The empirical performance of the proposed method is demonstrated in several different operational design domains through a series of cases covering a variety of fidelity levels (real-world and simulators), driving environments (highway, urban, and intersections), road users (car, truck, and pedestrian), and SV driving behaviors (human driver and self driving algorithms).

Vehicle-following behaviour in mixed traffic – role of lane position and adjacent vehicle

ABSTRACT In mixed traffic condition, varying vehicle dimensions and lack of lane discipline lead to parallel movement of vehicles in the same lane or between lanes. This results in a condition where the longitudinal response of vehicles gets affected by adjacent vehicles and their configurations. Correspondingly, the adjacent vehicle configurations are significantly influenced by the lane position of subject vehicle. To accommodate this scenario, the study formulates longitudinal acceleration models from trajectory data considering the influence of subject vehicle’s lane position along with adjacent vehicle characteristics. The response under different cases of subject vehicle’s lane position and adjacent vehicle configurations are evaluated. The statistical analysis indicates that the following behavior varies based on lane position of subject vehicle and adjacent vehicle attributes. It is also found that disregarding these attributes can produce significantly erroneous acceleration estimates. These features can improve existing following behavior models and can enhance the realism of the microscopic modeling scheme for mixed traffic conditions.

Design of a Secure Automated Driving Systems Test Data Interface

<div class="section abstract"><div class="htmlview paragraph">Vehicles equipped with Level 4 and 5 autonomy will need to be tested according to regulatory standards (or future revisions thereof) that vehicles with lower levels of autonomy are currently subject to. Today, dynamic Federal Motor Vehicle Safety Standards (FMVSS) tests are performed with human drivers and driving robots controlling the test vehicle’s steering wheel, throttle pedal, and brake pedal. However, many Level 4 and 5 vehicles will lack these traditional driver controls, so it will be impossible to control these vehicles using human drivers or traditional driving robots. Therefore, there is a need for an electronic interface that will allow engineers to send dynamic steering, speed, and brake commands to a vehicle. This paper describes the design and implementation of a market-ready Automated Driving Systems (ADS) Test Data Interface (TDI), a secure electronic control interface which aims to solve the challenges outlined above. The interface consists of a communication port integrated into automobiles which lack traditional manual controls. Via this interface dynamic test scenarios can be executed by plugging in a Vehicle Control Unit (VCU), which is an element of the TDI hardware/firmware intended to be available only to authorized users. Physically, the VCU can interface with the On-board Diagnostics (OBD) OBD-II connector that is present on essentially all modern automobiles. This paper also describes a demonstration of the complete VCU with the secure TDI using a vehicle equipped with a Dataspeed drive-by-wire kit, with its steering, throttle, and brake control systems mimicking the behavior of an autonomous subject vehicle.</div></div>

Traffic conflict assessment using macroscopic traffic flow variables: A novel framework for real-time applications

The present study develops a comprehensive traffic conflict assessment framework using macroscopic traffic state variables. To this end, vehicular trajectories extracted for a midblock section of a ten-lane divided Western Urban Expressway in India are used. A macroscopic indicator termed “time spent in conflict (TSC)” is adopted to evaluate traffic conflicts. The proportion of Stopping distance (PSD) is adopted as a suitable traffic conflict indicator. Vehicle-to-vehicle interactions in a traffic stream are two-dimensional, highlighting that the vehicles interact simultaneously in lateral and longitudinal dimensions. Therefore, a two-dimensional framework based on the influence zone of the subject vehicle is proposed and employed to evaluate TSCs. The TSCs are modeled as a function of macroscopic traffic flow variables, namely, traffic density, speed, the standard deviation in speed, and traffic composition, under a two-step modeling framework. In the first step, the TSCs are modeled using a grouped random parameter Tobit (GRP-Tobit) model. In the second step, data-driven machine learning models are employed to model TSCs. The results revealed that intermediately congested traffic flow conditions are critical for traffic safety. Furthermore, macroscopic traffic variables positively influence the value of TSC, highlighting that the TSC increases with an increase in the value of any independent variable. Among different machine learning models, the random forest (RF) model was observed as the best-fitted model to predict TSC based on macroscopic traffic variables. The developed machine learning model facilitates traffic safety monitoring in real-time.

Predictive trajectory planning for autonomous vehicles at intersections using reinforcement learning

In this work we put forward a predictive trajectory planning framework to help autonomous vehicles plan future trajectories. We develop a partially observable Markov decision process (POMDP) to model this sequential decision making problem, and a deep reinforcement learning solution methodology to learn high-quality policies. The POMDP model utilizes driving scenarios, condensed into graphs, as inputs. More specifically, an input graph contains information on the history trajectory of the subject vehicle, predicted trajectories of other agents in the scene (e.g., other vehicles, pedestrians, and cyclists), as well as predicted risk levels posed by surrounding vehicles to devise safe, comfortable, and energy-efficient trajectories for the subject vehicle to follow. In order to obtain sufficient driving scenarios to use as training data, we propose a simulation framework to generate socially acceptable driving scenarios using a real world autonomous vehicle dataset. The simulation framework utilizes Bayesian Gaussian mixture models to learn trajectory patterns of different agent types, and Gibbs sampling to ensure that the distribution of simulated scenarios matches that of the real-world dataset collected by an autonomous fleet. We evaluate the proposed work in two complex urban driving environments: a non-signalized T-junction and a non-signalized lane merge intersection. Both environments provide vastly more complex driving scenarios compared to a highway driving environment, which has been mostly the focus of previous studies. The framework demonstrates promising performance for planning horizons as long as five seconds. We compare safety, comfort, and energy efficiency of the planned trajectories against human-driven trajectories in both experimental driving environments, and demonstrate that it outperforms human-driven trajectories in a statistically significant fashion in all aspects.

Analysing freeway diverging risks using high-resolution trajectory data based on conflict prediction models

Abstract This study aims to develop a reliable safety evaluation model for diverging vehicles and investigates the impact of the surrounding traffic environment on freeway diverging risks. High-resolution trajectory data from three sites in the Netherlands (Delft, Ter-Heide and Zonzeel) were employed for the risk analysis. Linear regression (LR), support vector machine (SVM), random forest (RF), extreme randomize trees (ET), adaptive boosting (Adaboost), extreme gradient boosting (XGboost) and multilayer perceptron (MLP), were developed for safety evaluation. The result showed that MLP outperforms the other models for diverging risk prediction over all the indicators, conflict thresholds and locations. Pairwise matrix, shapely addictive explanation (SHAP), and LR algorithms were further adopted to interpret the influence of the surrounding environment. It indicates that an increase in traffic density, subject vehicle (SV) lateral speed, the distance of SV from ramp nose and SV length would increase the diverging risk. At the same time, an increase in leading vehicle speed and space headway would decrease diverging risk. Finally, spatial analysis was also conducted to explore the stability of identified traffic features regarding the impact on the diverging risk across the sites.