Autonomous Vehicle Environment Research Articles

Global point cloud registration network for large transformations

Three-dimensional registration is an established yet challenging problem that is key in many different applications, such as mapping the environment for autonomous vehicles, or modeling people for avatar creation, among others. Registration refers to the process of mapping multiple data into the same coordinate system by means of matching correspondences and transformation estimation. Novel proposals exploit the benefits of deep learning architectures for this purpose, as they learn the best features for the data, providing better matches and hence results. However, the state of the art is usually focused on cases of relatively small transformations, although in certain applications and in a real and practical environment, large transformations are very common. In this paper, we present ReLaTo (Registration for Large Transformations), an architecture that addresses the cases where large transformations happen while maintaining good performance for local transformations. This proposal uses a novel Softmax pooling layer to find correspondences in a bilateral consensus manner between two point sets, sampling the most confident matches. These matches estimate a coarse and global registration using weighted Singular Value Decomposition (SVD). A target-guided denoising step is applied to both the obtained matches and latent features to estimate the final fine registration considering the local geometry. All these steps are carried out following an end-to-end approach, which has been shown to perform better than 7 state-of-the-art registration methods in two datasets commonly used for this task (ModelNet40 and the Karlsruhe Institute of Technology and Toyota Technological Institute dataset, KITTI), especially in the case of large transformations.Graphic abstract

A NOVEL HYBRID DEEP LEARNING APPROACH FOR 3D OBJECT DETECTION AND TRACKING IN AUTONOMOUS DRIVING

Recently Object detection and tracking using fusion of LiDAR and RGB camera for the autonomous vehicle environment is a challenging task. The existingworks initiates several object detection and tracking frameworks using ArtificialIntelligence (AI) algorithms. However, they were limited with high false positives and computation time issues thus lacking the performance of autonomousdriving environment. The existing issues are resolved by proposing HybridDeep Learning based Multi Object Detection and Tracking (HDL-MODT) using sensor fusion methods. The proposed work performs fusion of solid stateLiDAR, Pseudo LiDAR, and RGB camera for improving detection and trackingquality. At first, the multi-stage preprocessing is done in which noise removal isperformed using Adaptive Fuzzy Filter (A-Fuzzy). The pre-processed fused image is then provided for instance segmentation to reduce the classification andtracking complexity. For that, the proposed work adopts Lightweight GeneralAdversarial Networks (LGAN). The segmented image is provided for objectdetection and tracking using HDL. For reducing the complexity, the proposedwork utilized VGG-16 for feature extraction which forms the feature vectors.The features vectors are then provided for object detection using YOLOv4.Finally, the detected objects were tracked using Improved Unscented KalmanFilter (IUKF) and mapping the vehicles using time based mapping by considering their RFID, velocity, location, dimension and unique ID. The simulation ofthe proposed work is carried out using MATLAB R2020a simulation tool andperformance of the proposed work is compared with several metrics that showthat the proposed work outperforms than the existing works.

Intention to adopt autonomous vehicles in China: a comparative study among residents in different-sized cities

ABSTRACT Transitioning to a fully Level-5 autonomous vehicle (AV) environment presents numerous challenges, notably influenced by public adoption intention. Previous studies have shown limitations in scope, population, and methodology. This study expands the Technology Acceptance Model to investigate AVs adoption intention across various city sizes in China. Through surveys in China, 2,662 responses were gathered in 2021 from mega, large, and small-to-medium cities. Using Multiple Indicators and Multiple Causes models, the study examines influencing factors on AVs adoption intention and population heterogeneities. Key findings emphasize the importance of adoption attitude, information provision, and perceived AV usefulness. Additionally, the impact of financial incentives, convenience, and several other factors varies across city sizes. The insights gained from the study can be utilized to develop more cost-effective policies and strategies tailored to different subgroups of the population to fully utilize the potential benefits of AVs while minimizing unintended consequences across diverse urban settings.

Pedestrians’ intent to cross in a fully autonomous vehicle environment; looking at crossing opportunity, eHMI message and wait time

Objectives Currently, pedestrians’ road-crossing decisions depend on the traffic at the crossing point, crossing opportunities, and circumstantial elements. Longer wait times on the curb and time pressure raise the number of traffic violations among pedestrians. The era of fully autonomous vehicles (FAVs) promises new interactions. External Human-Machine Interfaces (eHMIs) aim to increase the understanding of FAVs’ intentions and will influence pedestrians’ intent to cross. Yet, how pedestrian behaviors will change in the FAV era remains unclear. Further, there are no good metrics for evaluating the intricate interaction of crossing-related factors. Methods In a laboratory study, sixty participants observed crossing scenarios of a typical one-lane road with 10–12 FAVs driving along, two possible eHMI messages (“Cross!”/“Stop!”), and varying crossing opportunities (safe, risky, and unsafe). Participants had to indicate when they intended to cross the road using a designated button. Half of the participants were induced to feel time pressure. We developed index scores to convey the complexity of such crossing situations from the perspective of a pedestrian and used them to analyze the experimental data. Results In total, 48% of participants’ indications were to cross in safe crossing opportunities, 34% risky, and 18% unsafe; 69% were made when the eHMI proposition was “Cross!” and 31% for “Stop!”. High, significant, and negative correlations were found for compatible responses, e.g., with a safe crossing gap and an eHMI sign to “Cross!”. The opposite was true for incompatibility, e.g., a safe crossing gap and an eHMI sign to “Stop!” or an unsafe crossing gap and an eHMI sign to “Cross!”. The time pressure group’s median times to cross were shorter, making riskier decisions. Conclusion Our findings reflect a tendency to comply with the eHMI. In more complex scenarios where the crossing opportunity was uncertain or waiting times were longer, we observed that pedestrians were more inclined to follow the eHMI’s guidance to cross. These findings suggest that pedestrians may be more likely to follow the eHMI’s suggestion to cross than their self-skill, especially under time pressure. Our indices and metrics reflect the intricate interaction aspects of pedestrian behavior with eHMIs.

Employing cross-domain modelings for robust object detection in dynamic environment of autonomous vehicles

Employing cross-domain modelings for robust object detection in dynamic environment of autonomous vehicles

A Deep Learning Approach to Semantic Segmentation of Steel Microstructures

The utilization of convolutional neural networks (CNNs) for semantic segmentation has proven to be successful in various applications, such as autonomous vehicle environment analysis, medical imaging, and satellite imagery. In this study, we investigate the application of different segmentation networks, including Deeplabv3+, U-Net, and SegNet, each recognized for their effectiveness in semantic segmentation tasks. Additionally, in the case of Deeplabv3+, we leverage the use of pre-trained ResNet50, ResNet18 and MobileNetv2 as feature extractors for a comprehensive analysis of steel microstructures. Our specific focus is on distinguishing perlite and ferrite phases in micrographs of low-carbon steel specimens subjected to annealing heat treatment. The micrographs obtained using an optical microscope are manually segmented. Preprocessing techniques are then applied to create a dataset for building a supervised learning model. In the results section, we discuss in detail the performance of the obtained models and the metrics used. The models achieve a remarkable 95% to 98% accuracy in correctly labeling pixels for each phase. This underscores the effectiveness of our approach in differentiating perlite and ferrite phases within steel microstructures.

An Autoregressive Model of Electromagnetic Disturbances in An Autonomous Electric Vehicle’s Route

Electromagnetic Interference (EMI) can cause a malfunction of on-board electronic circuits in an autonomous electric vehicle and supporting electronic devices located in the environment of autonomous electric vehicles as well. In order to navigate an autonomous electric vehicle safely, it is important to have electromagnetic field characteristic in the environment. Since the information of electromagnetic field characteristic is hard to find, it needs to be modeled. This paper presents a model of electromagnetic field characteristic that is generated by using autoregression in order to estimate potential EMI. The EMI estimation is based on electromagnetic characteristic in an environment. Unlike other applications that use time history of data to build a model, we present a spatial electromagnetic field strength data in a previous route to estimate the future data in a new route. To obtain historical data for auto-regression process, we measured electric field strengths along a circular route in a campus near Jakarta. This surrounding environment represents a typical area of suburbs. The input variables for auto-regression process are the first 27 correlated data of 155 measured data. The result shows that the use of 13 predictor coefficient produces a variance of prediction error near to zero, with an improvement from maximum prediction error of 15.1257 to prediction error of 0.1862.

Leveraging LiDAR-Based Simulations to Quantify the Complexity of the Static Environment for Autonomous Vehicles in Rural Settings.

This paper uses virtual simulations to examine the interaction between autonomous vehicles (AVs) and their surrounding environment. A framework was developed to estimate the environment's complexity by calculating the real-time data processing requirements for AVs to navigate effectively. The VISTA simulator was used to synthesize viewpoints to replicate the captured environment accurately. With an emphasis on static physical features, roadways were dissected into relevant road features (RRFs) and full environment (FE) to study the impact of roadside features on the scene complexity and demonstrate the gravity of wildlife-vehicle collisions (WVCs) on AVs. The results indicate that roadside features substantially increase environmental complexity by up to 400%. Increasing a single lane to the road was observed to increase the processing requirements by 12.3-16.5%. Crest vertical curves decrease data rates due to occlusion challenges, with a reported average of 4.2% data loss, while sag curves can increase the complexity by 7%. In horizontal curves, roadside occlusion contributed to severe loss in road information, leading to a decrease in data rate requirements by as much as 19%. As for weather conditions, heavy rain increased the AV's processing demands by a staggering 240% when compared to normal weather conditions. AV developers and government agencies can exploit the findings of this study to better tailor AV designs and meet the necessary infrastructure requirements.

A multi-agent reinforcement learning-based longitudinal and lateral control of CAVs to improve traffic efficiency in a mandatory lane change scenario

Bottleneck areas are prone to severe traffic congestion due to the sudden drop in capacity. To improve traffic efficiency in the bottleneck area, this paper proposes a multi-agent deep reinforcement learning framework integrating collision avoidance strategies to improve traffic efficiency in a mandatory lane change scenario. The proposed method considers distance-keeping and lane-changing coordination in a connected autonomous vehicle (CAV) environment, by controlling vehicles' longitudinal and lateral movement to effectively reduce traffic congestion in a mandatory lane change scenario. This framework was trained and tested in a simulation environment that is the same as the natural driving environment. Compared with real-world data and the benchmark model (a Dueling Double Deep Q-Network-based model), the proposed model shows better performance in terms of average speed, travel time, throughput, and safety in the bottleneck area. The results show that the proposed model can effectively reduce traffic congestion and improve traffic efficiency in a mandatory lane change scenario.

A Cooperative Lane-Changing Strategy for Weaving Sections of Urban Expressway under the Connected Autonomous Vehicle Environment

To alleviate the lane-changing conflicts between weaving vehicles and enhance the traffic efficiency in the weaving section of urban expressway under the connected autonomous vehicle (CAV) environment, a cooperative lane-changing strategy for CAVs is proposed. The strategy consists of an upper layer of decision making, which determines the lane-changing sequences of weaving vehicles based on their lane-changing advantages quantified by a set of utility functions, and a lower layer of control, which generates detailed instructions of speed adjustments and lane-changing manoeuvres for weaving vehicles. To verify the effectiveness of the proposed strategy under different traffic demand settings, a numerical simulation, including a base case and a control case, is conducted. Then, to further verify the effectiveness of the proposed strategy for the mixed traffic state and compare its performance with the existing CAV lane-changing method, benchmark and comparison tests with six different market penetration rates (MPRs) of CAVs are carried out under the congested demand setting. In addition, the delay improvement ratio, inverse time-to-collision, and ratio of large deceleration time are selected as performance indicators to investigate the effect of the proposed strategy on enhancing the operational efficiency, traffic safety, and passenger’s comfort within the weaving section. According to the simulation results, the overall efficiency, safety and comfort in the weaving section under the CAV environment, are all improved, when the proposed strategy is applied to weaving vehicles. The proposed strategy is also superior to the existing CAV lane-changing method on maintaining traffic efficiency and safety. Therefore, the proposed cooperative lane-changing strategy, based on CAV technologies, shows good potential in solving the problem of lane-changing conflicts within the weaving section and facilitating the traffic management and traffic control of urban expressway.

Investigating Residential Built Environment Effects on Rank-Based Modal Preferences and Auto-Ownership

Studies in the residential self-selection literature, which use attitudinal factors to model the jointness of residential and travel-related choices, assume a unidirectional impact from attitudes to behavior; however, such an assumption may be violated under several circumstances. In this current study, we allow the error terms of the attitudinal factors to be correlated with the main outcomes as we jointly model residential choice, auto-ownership level, and rank-based modal preferences. In our joint model, we use green lifestyle propensity and luxury lifestyle propensity as the two stochastic latent constructs. The empirical data for this study is drawn from the 2019 multi-city Transformative Technologies in Transportation survey for the city of Austin that elicited information with respect to individuals’ residential location, auto-ownership, and modal preferences through a stated preference experiment in a futuristic autonomous vehicle environment. Results indicate significant unobserved correlations between the latent constructs and the main outcomes; ignoring such endogeneity leads to underestimations of the “true” causal effect of high-density neighborhood (HDN) living on travel-related choices, which can have consequences for policy-making. In our analysis, the “true” causal effect of HDN living on auto-ownership suggests that, on average, the auto-ownership level would reduce by about 29% when an individual is shifted from a non-HDN to a HDN. Furthermore, the probability of using the bicycle mode for non-work pursuits is estimated to increase by 8%, and that of using a private vehicle is estimated to decrease by 3.1%, when individuals are moved from a non-HDN to a HDN.

Optimal Differential Variable Speed Limit Control in a Connected and Autonomous Vehicle Environment for Freeway Off-Ramp Bottlenecks

When traffic congestion occurs on freeway off-ramp bottlenecks, the traffic state becomes complicated and changeable, which leads to increased vehicle travel time and decreased traffic safety and traffic efficiency. Variable speed limit (VSL) control is an effective method to improve traffic conditions, increase bottleneck throughput, improve traffic efficiency, and reduce emissions. Currently, there is an emerging trend of using connected and autonomous vehicle (CAV) technology to develop VSL control. This paper proposes an optimal differential variable speed limit (DVSL) control strategy under mixed CAVs and human-driven vehicles (HVs) environment for freeway off-ramp bottlenecks. The proposed DVSL control considers the characteristics of on-ramp, off-ramp, and mixed traffic flow (i.e., CAVs coexist with HVs). The proposed optimal DVSL control can describe and forecast the dynamics of traffic flow, and can set different speed limits across each lane with a multiple-objective function of total travel time (TTT) and total travel distance (TTD). A model predictive control (MPC) approach was utilized to optimize the DVSL control algorithm. The designed DVSL control was tested on a real-word freeway section with a simulated off-ramp bottleneck. The simulation results show that the proposed control strategy outperforms other existing methods in terms of improving the mobility of a freeway off-ramp bottleneck and maximizing the environmental benefits. Sensitivity analysis shows that the proposed control strategy can improve performance with the increase of the penetration rate (PR) of CAVs. The proposed methods form the basis of VSL control at off-ramp sections under mixed traffic environment.

Equilibrium analysis of morning commuting and parking under spatial capacity allocation in the autonomous vehicle environment

This study analytically investigates the morning commuting and parking patterns of autonomous vehicles (AVs) under different spatial road capacity allocation schemes (i.e., capacity split between inbound and outbound travel directions). Given that self-driving AV might park far away from commuters’ destination, we investigate equilibrium departure/arrival and parking patterns for AVs subject to the spatial road capacity allocation. We also analyse the system optimum traffic pattern for AV morning commute under a given capacity allocation scheme. Furthermore, we examine optimal capacity allocation strategies under user equilibrium and system optimum AV traffic patterns, respectively, which aim to minimise the total system travel cost. Numerical studies are conducted to illustrate the model and analysis. The results reveal the sensitivity of different efficiency metrics with respect to AV parking supply and road capacity allocation schemes, and provide insights into the infrastructure management with future automated transport.

Learning Configurations of Operating Environment of Autonomous Vehicles to Maximize their Collisions

Autonomous vehicles must operate safely in their dynamic and continuously-changing environment. However, the operating environment of an autonomous vehicle is complicated and full of various types of uncertainties. Additionally, the operating environment has many configurations, including static and dynamic obstacles with which an autonomous vehicle must avoid collisions. Though various approaches targeting environment configuration for autonomous vehicles have shown promising results, their effectiveness in dealing with a continuous-changing environment is limited. Thus, it is essential to learn realistic environment configurations of continuously-changing environment, under which an autonomous vehicle should be tested regarding its ability to avoid collisions. Featured with agents dynamically interacting with the environment, Reinforcement Learning (RL) has shown great potential in dealing with complicated problems requiring adapting to the environment. To this end, we present an RL-based environment configuration learning approach, i.e., <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepCollision</i> , which intelligently learns environment configurations that lead an autonomous vehicle to crash. DeepCollision employs Deep Q-Learning as the RL solution, and selects <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">collision probability</i> as the safety measure, to construct the reward function. We trained four DeepCollision models and conducted an experiment to compare them with two baselines, i.e., random and greedy. Results show that DeepCollision demonstrated significantly better effectiveness in generating collisions compared with the baselines. We also provide recommendations on configuring DeepCollision with the most suitable time interval based on different road structures.

A macroscopic dynamic network loading model using variational theory in a connected and autonomous vehicle environment

Recent studies on macroscopic fundamental diagram (MFD) have shifted towards enhancing MFD dynamics in terms of efficiency and realism. In this paper, a multi-reservoir dynamic network loading (MRDNL) model for a large-scale urban road network is developed. The proposed framework consists of a hyper-link and a hyper-node model. The hyper-link model is constructed using variational theory (VT) by introducing a set of constraints that construct an upper bound for upstream and downstream cumulative accumulations. Using VT, LWR is incorporated into MFD dynamics to enhance traffic propagation at reservoirs. The hyper-link model captures important traffic phenomena, such as kinematic waves, queueing, and congestion, while the hyper-node model functions as a medium for transferring flow to other parts of the multi-reservoir urban network. A numerical scheme is advanced for solving the proposed MRDNL model and its performance is evaluated using a hypothetical traffic network. Compared to previous MFD-based dynamic network loading models, our approach is more computationally efficient since it does not require discretization and offers more realistic solutions by considering multiple kinematic waves and including the dynamics of urban traffic signals. This study, further, incorporated connected and autonomous vehicles (CAVs) into MFD dynamics and evaluated the network-wide effect of introduction of CAVs in urban traffic networks. For a network with 100% CAV market penetration rate, an approximate enhancement of 30% in total network outflow was found.

Managing the Morning Commute Problem with Tradable Credit Schemes under a Fully Autonomous Vehicle Environment

This study investigates the morning commute problem in a single corridor with a bottleneck under a fully automated vehicle environment. By extending Vickrey’s bottleneck model, we formulate the joint decisions on departure time and parking choices of morning commuters who make trade-offs among travel cost, the cost associated with parking, and the cost of tradable credits. To alleviate traffic congestion and improve social welfare, we propose a time-varying tradable credit scheme and integrate it with the morning commute problem. We explore the travel patterns and the optimal design of tradable credit schemes for the morning commute problem with homogeneous and heterogeneous commuters, respectively. For the homogeneous case, we derive the conditions on the tradable credit scheme to ensure the existence of equilibrium. The system-wide travel cost decreases with parking density after tradable credits are incorporated. Additionally, the efficiency of the tradable credits scheme can be improved by increasing the rate of credit charge rate. For the heterogeneous case, we propose an initial distribution strategy and combine it with the optimal tradable credit in order to guarantee social equity. The commuters with a low value of time (VOT) should be allocated more credits and the commuters with high VOT should be charged more credits. For both cases at system optimum, we find that the equilibrium price of tradable credits increases with parking density and decreases with the total amount of tradable credits.

Optimal Dynamic Supply of Parking Permits Under Uncertainties: A Stochastic Control Approach

This paper proposes an optimal dynamic supply of parking permits under uncertainties in the mixed traffic environment of regular vehicles (RVs) and autonomous vehicles (AVs). The uncertainties from queuing delay and cruising-for-parking time, respectively, are derived from parking permit holders’ unascertainable early arrival/late departure and the randomness of vehicle speed for cruising-for-parking users. We formulate the dynamic system optimum as a stochastic optimal control problem, aiming to minimize the total time lost for all parking users. On this basis, we express the optimality condition as Hamilton-Jacobi-Bellman equation and propose an optimal control strategy for dynamic supply of parking permits (called optimal parking supply scheme). The proposed scheme is determined by two observable and measurable variables, i.e., queuing time and cruising-for-parking time. We find that the scheme can significantly reduce the total system time lost, especially when the uncertainty from parking permit holders is at a high level or the percentage of AVs is small. Finally, a case study of a selected sample area in Beijing is conducted, which verifies the efficiency of the proposed scheme numerically.

Assessment of CNN-Based Models for Odometry Estimation Methods with LiDAR

The problem of simultaneous localization and mapping (SLAM) in mobile robotics currently remains a crucial issue to ensure the safety of autonomous vehicles’ navigation. One approach addressing the SLAM problem and odometry estimation has been through perception sensors, leading to V-SLAM and visual odometry solutions. Furthermore, for these purposes, computer vision approaches are quite widespread, but LiDAR is a more reliable technology for obstacles detection and its application could be broadened. However, in most cases, definitive results are not achieved, or they suffer from a high computational load that limits their operation in real time. Deep Learning techniques have proven their validity in many different fields, one of them being the perception of the environment of autonomous vehicles. This paper proposes an approach to address the estimation of the ego-vehicle positioning from 3D LiDAR data, taking advantage of the capabilities of a system based on Machine Learning models, analyzing possible limitations. Models have been used with two real datasets. Results provide the conclusion that CNN-based odometry could guarantee local consistency, whereas it loses accuracy due to cumulative errors in the evaluation of the global trajectory, so global consistency is not guaranteed.

EVCT: An Efficient VM Deployment Algorithm for a Software-Defined Data Center in a Connected and Autonomous Vehicle Environment

Software-defined data centers (SDDC) are an emerging softwarized model that can monitor the virtual machines’ allocation atop the cloud servers. SDDC consists of softwarized entities like Virtual Machine (VM) and hardware entities like servers and connected switches. SDDCs apply VM deployment algorithms to preserve efficient placement and processing data traffic generated from the Connected and Autonomous Vehicles (CAV). To enhance user satisfaction, SDDC providers are always looking for an intellectual model to monitor large-scale incoming traffics, such as the Internet of Things (IoT) and CAV applications, by optimizing service quality and service level agreement (SLA). This paper is motivated by this, raising an energy-efficient VM cluster placement algorithm named <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">EVCT</i> to handle service quality and SLA issues in an SDDC in a CAV environment. EVCT algorithm leverages the similarity between VMs and models the problem of VM deployment into a weighted directed graph. Based on the amount of traffic between VM, EVCT adopts the “maximum flow and minimum cut theory” to cut the directed graph and achieve high energy-efficient placement for VMs. The proposed algorithm can efficiently reduce the energy consumption cost, provide a high quality of services (QoS) to users, and have good scalability for the variable workload. We have also carried out a series of experiments to use the real-world workload to evaluate the performance of the EVCT. The results illustrate that the EVCT surpasses the state-of-the-art algorithms in terms of energy consumption cost and efficiency.

A Hybrid Method for Dealing With DVL Faults of SINS/DVL Integrated Navigation System

Aiming at the situation that the underwater integrated navigation system is easy to be disturbed and the output is unavailable during the task of underwater vehicle, a RVBAKF/ LSTM hybrid method is proposed in this paper. Considering the actual underwater working environment of autonomous underwater vehicles (AUV), the measured value of Doppler velocity log (DVL) is prone to outliers or even interruption, so it is very important to improve the fault tolerance of integrated navigation system. For the processing of outliers, a robust adaptive filter based on VB theory is proposed. Its advantage is to realize the adaptation of time-varying noise and reduce the interference of outliers to the integrated navigation system. Firstly, selecting the SINS output related to AUV velocity, such as velocity, attitude, angular velocity and specific force as the training input when the DVL output is normal. When DVL output is interrupted, the long short-term memory (LSTM) model is used to provide pseudo measurement values for integrated navigation system. The effectiveness of the method is verified by the on-board experimental data. When the DVL output is interrupted, the method can effectively improve the accuracy of the integrated navigation system. The method in this paper is compared with four different hybrid methods, and the results are obviously better than the four methods.