Impacts of Automated Vehicles on Highway Infrastructure
Paul Carlson, Road Infrastructure, Inc.Show Abstract
This paper summarizes the results of a multiphase federally-funded research effort, which involved a comprehensive literature review, engagement with highway infrastructure owners and operators, and interviews with industry experts and key stakeholders to document the potential impact of automated vehicles (AVs) on highway infrastructure. The research attempted to identify the state of the practice among highway infrastructure owners/operators, any gaps in knowledge, and agency preparedness levels regarding the impacts of AV use on highway infrastructure. The operations or policy aspects of the AV infrastructure impacts were not covered. Four different categories of the physical roadway infrastructure were considered for this study, including the uniformity and quality of traffic control devices, the changing demands of intelligent transportations systems (ITS) devices, structural requirements for pavements and bridges, the effects on multimodal infrastructure (bike lanes and Complete Streets designs), and the potential need for other roadside infrastructure (e.g., guardrails and communications systems for roadway digital infrastructure). The research identified early strategies for design and maintenance of certain physical highway infrastructure elements that are believed to benefit AVs as well as human drivers.
Cooperative Car-Following and Merging: A Novel Merge Control Strategy Considering Cooperative Adaptive Cruise Control and Courtesy
Yuanchang Xie (firstname.lastname@example.org), University of Massachusetts, LowellShow Abstract
This study focuses on how to improve the merge control prior to lane reduction points due to either accidents or constructions. A Cooperative Car-following and Merging (CCM) control strategy is proposed considering the coexistence of Automated Vehicles (AVs) and Human-Driven Vehicles (HDVs). CCM introduces a modified/generalized Cooperative Adaptive Cruise Control (CACC) for vehicle longitudinal control prior to lane reduction points. It also takes courtesy into account to ensure that AVs behave responsibly and ethically. CCM is evaluated using microscopic traffic simulation and compared with no control and CACC merge strategies. The results show that CCM consistently generates the lowest delays and highest throughputs approaching the theoretical capacity. Its safety benefits are also found to be significant based on vehicle trajectories and density maps. AVs in this study does not need to be fully automated and can be at Level-1 automation. CCM only requires automated longitudinal control and information sharing among vehicles such as CACC, and CACC is already commercially available on many new vehicles. Also, it does not need 100% CACC penetration, presenting itself as a promising and practical solution for improving traffic operations in lane reduction transition areas such as highway work zones.
Better than ‘Average’: Potential Crash Rate Standards for Automated Vehicles
Noah Goodall (Noah.Goodall@VDOT.Virginia.gov), Virginia Department of TransportationShow Abstract
Most automobile manufacturers and several technology companies are testing automated vehicles on public roads. While automation of the driving task is expected to reduce crashes, there is no consensus regarding how safe an automated vehicle must be before it can be deployed. An automated vehicle should be at least as safe as the average driver, but national crash rates include drunk and distracted driving, meaning that an automated vehicle that crashes at the average rate is somewhere between drunk and sober. In this paper, automated vehicle safety standards are explored from three perspectives. First, crash rates from naturalistic driving studies are used to determine the crash risk of the model (i.e. sober, rested, attentive, cautious) driver. Second, stated preference surveys in the literature are reviewed to estimate the public’s acceptable automated vehicle risk. Third, crash, injury, and fatality rates from other transportation modes are compared as baseline safety levels. A range of potential safety standards is presented as a guide for policymakers, regulators, and automated vehicle developers to assist in validating the safety of automated driving technologies for public use.
Investigating the Potential of Truck Platooning on Energy Savings: an Empirical Study on the U.S. National Highway Freight Network
Xiaotong Sun, University of Michigan, Ann ArborShow Abstract
Yafeng Yin, University of Michigan, Ann Arbor
Bo Zou, University of Illinois, Chicago
Haochen Wu, University of Michigan
Mojtaba Abdolmaleki, University of Michigan
Truck platooning enabled by the connected automated vehicle (CAV) technology has been demonstrated to effectively reduce fuel consumption for trucks in a platoon. However, given the limited number of trucks in the traffic stream, it remains questionable how much energy savings it may yield for a practical freight system if we only rely on ad-hoc platooning. Assuming the presence of a central platooning coordinator, this paper emerges to substantiate truck-platooning benefits in fuel economy by exploiting platooning opportunities from the United States domestic truck demands on its highway freight network. An integer programming model is utilized to schedule trucks' itineraries to facilitate the formation of platoons at platoonable locations to maximize energy savings. A simplification of the real freight network and an approximation algorithm are performed to solve the model efficiently. By analyzing the numerical results obtained, this study quantifies the importance of scheduled platooning in improving trucks' fuel economy. Furthermore, the allowable platoon size, schedule flexibility, fuel efficiency all play a crucial role in energy savings. Specifically, by assuming that following vehicles in a platoon obtain a 10\% energy reduction, an average energy reduction of 8.48\% per truck can be achieved for the overall network if the maximum platoon size is seven, and the schedule flexibility is 30 minutes. The cost-benefit analysis provided in the end suggests that the energy-saving benefits can offset the investment cost on truck platooning technology.
A Learning-based Stochastic Driving Model for Autonomous Vehicle Testing
Lin Liu, University of Michigan, Ann ArborShow Abstract
Shuo Feng (email@example.com), University of Michigan, Ann Arbor
Yiheng Feng, Purdue University
Xichan Zhu, Tongji University
Henry Liu, University of Michigan, Ann Arbor
In the simulation-based testing and evaluation of autonomous vehicles (AVs), how background vehicles (BVs) drive directly influences the AV’s driving behavior and further impacts the testing result. Existing simulation platforms either use pre-determined trajectories or deterministic driving models to model the BVs’ behaviors. However, pre-determined BV trajectories can not react to the AV's maneuvers and deterministic models are different from real human drivers due to lake of stochastic components and errors. Both methods lead to unrealistic traffic scenarios. This paper presents a learning-based stochastic driving model that meets the unique needs of AV testing, i.e. interactive and human-like. The model is built based on the long-short-term-memory (LSTM) architecture. By incorporating quantile-regression to the loss function of the model, the stochastic behaviors are reproduced without any assumption or prior knowledge of human drivers. The model is trained with the naturalistic driving data (NDD) from the Safety Pilot Model Deployment (SPMD) project and compared with a modified Intelligent Driving Model (IDM). Analysis from individual trajectories shows that the proposed model can reproduce trajectories from human drivers much better than IDM. To validate the ability of the proposed model in generating a naturalistic driving environment, traffic simulation experiments are implemented. The results show that the traffic flow parameters such as speed, range, and headway distribution match closely with the NDD, which is of significant importance for AV testing and evaluation.
Corner Case Generation and Analysis for Safety Assessment of Autonomous Vehicles
Haowei Sun, University of MichiganShow Abstract
Shuo Feng (firstname.lastname@example.org), University of Michigan, Ann Arbor
Xintao Yan, University of Michigan
Henry Liu, University of Michigan, Ann Arbor
Testing and evaluation is a crucial step in the development and deployment of Connected and Automated Vehicles (CAVs). To comprehensively evaluate the performance of CAVs, it is of necessity to test the CAVs in safety-critical scenarios, which rarely happen in naturalistic driving environment. Therefore, how to purposely and systematically generate these corner cases becomes an important problem. Most existing studies focus on generating adversarial examples for perception systems of CAVs, whereas limited efforts have been put on the decision-making systems, which is the highlight of this paper. As the CAVs need to interact with numerous background vehicles (BVs) for a long duration, variables that define the corner cases are usually high dimensional, which makes the generation a challenging problem. In this paper, a unified framework is proposed to generate corner cases for the decision-making systems. To address the challenge brought by high dimensionality, the driving environment is formulated based on Markov Decision Process, and the deep reinforcement learning techniques are applied to learn the behavior policy of BVs. With the learned policy, BVs will behave and interact with the CAVs more aggressively, resulting in more corner cases. To further analyze the generated corner cases, the techniques of feature extraction and clustering are utilized. By selecting representative cases of each cluster and outliers, the valuable corner cases can be identified from the library of all generated corner cases. Simulation results of a highway driving environment show that the proposed methods can effectively generate and identify the valuable corner cases.
Cut-in Test Scenarios Generation Method for Autonomous Vehicles
Shuang Liu, Tongji UniversityShow Abstract
Xuesong Wang (email@example.com), Tongji University
It’s a critical step to generate test cases widely covering High-risk scenarios in the real traffic environment for Autonomous Vehicle (AV) simulation test. This study aims to design a method of AV test scenario construction based on Monte Carlo method and Importance Sampling method. Two thousand and forty-nine cut-in events were extracted from the Shanghai Naturalistic Driving Study (SH-NDS) data, the corresponding scenario information for each event was used as scenario parameters. Time To Collision is used as the evaluation index. The results show that test scenarios generated by Monte Carlo method cover a larger area, however, it cannot increase the proportion of dangerous scenarios and highly-risky scenarios. Monte Carlo method combined with Importance Sampling are capable of generating scenarios covering all risk levels. The number of dangerous scenarios generated by this method is 1.61 times more than that simply generated by Monte Carlo method, which can better support autonomous driving testing.
Cooperative Control of Automated Vehicles of Different Cooperation Classes at a Stop-Controlled Intersection
Saeid Soleimaniamiri, University of South FloridaShow Abstract
Handong Yao, University of South Florida
Amir Ghiasi, Leidos, Inc.
Xiaopeng Li (firstname.lastname@example.org), University of South Florida
Pavle Bujanovic, Federal Highway Administration (FHWA)
Govindarajan Vadakpat, Federal Highway Administration (FHWA)
Taylor Lochrane, Federal Highway Administration (FHWA)
The objective of this paper is to propose a cooperative control framework for Cooperative-Automated Driving System (C-ADS)-equipped vehicles at a stop-controlled intersection in the Transportation Systems Management and Operations (TSMO) context. The proposed framework has two main components: 1-Critical Time Step Estimation (CTSE), 2-Trajectory Smoothing (TS). First, the CTSE component estimates a set of critical time steps (e.g., stopping time at the stop line, entering time to the intersection box) for each C-ADS-equipped vehicle. Second, the TS component is called at each C-ADS-equipped vehicle in a decentralized manner to control C-ADS-equipped vehicle trajectory based on the estimated critical time steps and its own cooperation behavior. This cooperative control framework focuses the infrastructure system only on critical high-level scheduling decisions while leaving complex low-level trajectory control and collision avoidance to individual C-ADS-equipped vehicles in a decentralized manner. Thus, it much reduces operational complexity and associated risks and liabilities for traffic operators. Also, it distributes the computational burden among different entities in an edging computing structure and thus makes it much more suitable for real-time applications. Further, this study for the first time investigates different cooperation classes defined in the SAE J3216 standard for stop-controlled intersections. Simulation results show that all considered performance measures (e.g., throughput, energy consumption, travel delay, etc.) will be much improved and the backward shockwave propagation will be reduced as the cooperation class of C-ADS-equipped vehicles increases. This study produces fundamental algorithms to be implemented in the FHWA CARMA open-source software (OSS).
Machine Learning Aided Platoon-Based Cooperative Lane-change Control Using MPC Approach
hanyu zhang (email@example.com), University of FloridaShow Abstract
Lili Du, University of Florida
Jinglai Shen, University of Maryland, Baltimore County
This study develops a platoon-based cooperative lane-change control (PB-CLC), which coordinates the trajectories of a CAV platoon under platooning control to accommodate the lane-change requests from several subject CAVs on the adjacent lane, aiming to reduce the negative impacts of lane-change maneuvers on the platoon on the premise of ensuring individual CAV’s safety and mobility. Mathematically, the PB-CLC control is established by the MPC approach involving a mixed integer nonlinear programming optimizer (MINLP-MPC), which considers multiple objectives such as ensuring traffic smoothness, driving comfort and lane-change response promptness. A machine learning aided distributed branch and bound algorithm (ML-DBB) is developed to address the computation challenge of solving the MINLP-MPC quickly (< 1 second) to adapt this online application. Specifically, built upon computer simulation and c-LHS sampling, supervised machine learning models are developed offline to predict a reduced solution space of the integer variables, which is further integrated into the distributed branch and bound method to solve the MINLP efficiently online. Extensive numerical experiments validate the effectiveness and applicability of the ML-DBB algorithm for this study. Besides, our experiments based on the field data further confirm the effectiveness of the PB-CLC control for ensuring safe and prompt lane-change as well as traffic smoothness and efficiency.
Comparison of Automated Vehicle Struck-from-Behind Crash Rates with National Rates Using Naturalistic Data
Noah Goodall (Noah.Goodall@VDOT.Virginia.gov), Virginia Department of TransportationShow Abstract
Automated vehicle developers in California are required to submit records of crashes and distances traveled in autonomous mode for all vehicles in their fleets. Several studies have investigated this database to compare automated vehicle crash rates with national rates. Although automated vehicles are struck from behind in 73% of their autonomous mode crashes, this is the first study to compare automated vehicle struck-from-behind crash rates to national rates using equivalent crash definitions. Rear-end collisions have substantially public health and economic impacts, representing a third of a collisions and $3.9 B in annual economic costs. In this study, automated vehicles were found to be struck from behind while in autonomous mode 17.2 (14.2–20.7, 95% CI) times per million-miles traveled, significantly higher than human-driven vehicles in naturalistic driving studies (3.6, 3.0–4.3, 95% CI). These differences narrow when comparing urban driving and business/industrial driving in the naturalistic driving studies with AV testing in similar environments. Automated vehicles were more likely to be struck when stopped than when moving, suggesting that automated vehicles’ decisions about where and when to stop at intersection are more plausible as contributing factors than unexpected rates of deceleration.
Investigation on Factors that Influence People's Acceptance of Automated Driving – A Kentucky Case Study
Song Wang, University of LouisvilleShow Abstract
Richard Li (firstname.lastname@example.org), University of Louisville
Existing studies have been focusing on identifying contributing factors that significantly impact the acceptance of emerging automated driving technology from socio-demographic perspectives. There lacks a comprehensive exploration of why these contributing factors influence people’s acceptance of the emerging automated driving technology from the mechanism’s perspective, which Automated Vehicle (AV) policymakers can obtain a more thorough understanding of the general public’s overall opinion on this new technology. Therefore, by using the statewide survey data in Kentucky , this study aims to reveal contributing factors that significantly influence people's acceptance of the emerging automated driving technology via clustering, and structural equation modeling (SEM) approaches. Survey respondents were assessed about their general opinion of AV on a 5-level Likert scale from "very negative" to "very positive". In general, 52% of the Kentucky residents are either "positive" or "very positive" about AV technology. Only 21% are either "negative" or "very negative". From the perspective of geographical setting, 60.8% of urban respondents and 42.7% of rural respondents are either "positive" or "very positive", respectively. 5.3% of urban respondents and 10.1% of rural respondents are "very negative". The differences of counties/districts of Kentucky in terms of the acceptance level were geographically visualized. Clusters of urbanized areas have relatively higher AV acceptance levels compared to rural clusters. The final SEM model identified contributing factors that significantly influence the overall acceptance level. Analysis of the odds ratios of contributing factors indicates that rural setting reduces the probability of having a higher overall acceptance level by 24.5%.
Sharing the Road with Autonomous Vehicles: A Qualitative Analysis of the Perceptions of Pedestrians and Bicyclists
Md Tawhidur Rahman, West Virginia UniversityShow Abstract
Kakan Dey (email@example.com), West Virginia University
Subasish Das, Texas A&M University
Melissa Sherfinski, West Virginia University
Public perception is considered an important metric to develop a better understanding of the acceptance of autonomous vehicles (AVs). This study investigated how pedestrians and bicyclists perceived AVs, applying a combined inductive and deductive data analysis approach. Survey responses of pedestrians and bicyclists in Pittsburgh, Pennsylvania, USA, collected by Bike Pittsburgh (Bike PGH) in 2019, were analyzed in this research. AVs following traffic rules appropriately and AVs driving safer than the human drivers were the most notable positive perceptions towards AVs. Pedestrians and bicyclists showed comparatively fewer negative perceptions towards AVs than positive perceptions. Negative perceptions mostly included a lack of perceived safety and comfort around AVs and trust in the AV technology. Respondents also concerned about AV technology issues (e.g., slow and defensive driving, disruptive maneuver), while sharing the road with AVs. Their views on AV safety significantly influenced perceptions of the respondents, familiarity with the technology, the extent respondents followed AV on the news, and household automobile ownership. Regulating AV movement on roadways, developing safety assessment guidelines, and controlling oversights of improper practices by AV companies were the major suggestions from the survey participants. The findings of this study might help AV companies to identify potential improvement needed in AV technology to increase acceptance of pedestrians and bicyclists, and policymakers to develop policy guidelines to ensure safe road sharing among pedestrians, bicyclists, and AVs.
Qualifying the Driving Environment Dynamics from the View of Autonomous Vehicles (AVs): Inspiration for Qualitatively Defining Operational Design Domains (ODDs)
Xing Fu, University of AlabamaShow Abstract
Jun Liu (firstname.lastname@example.org), University of Alabama
Alexander Hainen, University of Alabama
Asad J. Khattak, University of Tennessee
Until the Level 5 autonomous vehicle (AV) becomes a reality, automated driving systems (ADS) are expected to operate only under special environments without safety driver. A taxonomy - Operational Design Domain (ODD) has been defined to describe the conditions under which a given ADS is designed to function. The ODDs are currently described in a qualitative way, such as freeway, light rain, minimal traffic, etc. However, the ADS may prefer quantitative information feeds to determine whether a vehicle is within its ADS’s ODD. Using the data from Lyft, this study is to provide a quantitative method to depict the dynamic driving environments from the view of an AV, and to offer implications for qualitatively defining ODDs for ADS. From the camera and LIDAR data, objects and their relationships with the host vehicle or ego car were identified and the relationships were measured by metrics proposed in this study. The dynamics of driving environments were mapped into the road network, as a historical reference of the traffic dynamics for ADS. This study built a logit model to relate the ego car’s instantaneous maneuvers to the metrics that are proposed to quantitatively measure driving environments. The model revealed that the metrics are significantly related to vehicle maneuvers. It implies that these metrics may be suitable for quantifying ODDs for ADS. The quantified driving environments based on historical sensor data could be integrated into a sematic map for ADS to use as a reference to determine whether a driving environment meets its ODD. The public agencies may also use the semantic map to improve their environment to meet the ODD requirements by ADS.
Autonomous Vehicles’ Car-following Drivability Evaluation Based on Driving Behavior Spectrum Reference Model
Xiao Qi, Tongji UniversityShow Abstract
Ying Ni, Tongji University
Yiming Xu, University of Florida
Ye Tian, Tongji University
Junhua Wang, Tongji University
Jian Sun, Tongji University
As per existing reports, most of the accidents with Autonomous Vehicles (AVs) are not only caused by the functionality of AVs’ own autonomous system, but rather by the fact that human driving vehicles (HV) cannot understand AVs’ driving behavior properly. Such misunderstanding leads to danger situations during interaction. However, few research considered this kind of social HV-AV interactive safety into evaluation processes. Therefore, to evaluate the difference of driving behaviors between AVs and HVs is necessary. DRIVABILITY is defined to describe the driving capability of AVs in the HV-AV mixed traffic flow in this paper. How to determine the drivability evaluation indicator and to evaluate the difference of drivability through the quantitative method are challenges for drivability evaluation. Thus, a driving behavior spectrum reference model is proposed to evaluate AVs’ car-following drivability. Intellegent Driver Model (IDM) is used as the surrogate models to describe the different human drivers’ car-following behavior. An evaluation indicator called desired reaction time (DRT) is proposed to reflect the car-following drivability comprehensively. Relative entropy and cumulative frequencies are used to quantify the differences of driving behavior and to establish driving behavior spectrum. The proposed method, based on human naturalistic driving data, has been used to establish a human driving behavior spectrum as a reference to evaluate the three kinds of AVs’ car-following drivability in simulation software VTD. The result shows that the drivability of the default AV, the comfortable AV and the brisk AV are the 35th, 8th and 55th percentile of human drivers, respectively.
Reliability Analysis of a Passing Warning System on Two-Lane Highways Using Driving Simulator Data
Udai Hassein, Ryerson UniversityShow Abstract
Passing maneuvers are considered an effective measure to improve mobility levels along two-lane highways. Passing collision warning systems (PCWS) can help drivers avoid passing collisions two-lane highways by reducing the chance of human error. This paper analyses the reliability of a PCWS on passing maneuvers and driver behaviour in response to a passing warning system. This paper presents reliability models of the PSD No-Warning and PSD Warning that accounts for the variability of the input random variables to offer a better representation of real-life conditions. The reliability-based PSD models were developed using the First-Order Second-Moments (FOSM) method and a Monte Carlo Simulation (MCS) was used to validate the models. The proposed PCWS uses a radar sensor placed in the passing vehicle to detect opposing vehicles travelling in the left lane and calculate their relative distance and speed in order to estimate the time to collision. The proposed models account for the variability in the parameters by using the mean and standard deviation in a closed form estimation method. The analysis was performed, and the corresponding PSD No-Warning and PSD Warning distribution was established using empirical data obtained in a driving simulator carried out in Toronto, Canada. A comparison of the results of the proposed models, which reflect driver behaviour, and those of existing models is presented. The results indicate that the warning systems are beneficial and that they provide the driver with a certain level of comfort.
Detecting and Rectifying Vehicle Malicious Misbehavior for Intersection Movement Assist: Sensor-Based Misbehavior Detection Study
Boon Teck Ong, Booz Allen Hamilton, Inc.Show Abstract
Joshua Kolleda, Booz Allen Hamilton, Inc.
Saleh Mousa, Texas Department of Transportation
Scott Andrews, Cogenia Partners, LLC
Dennis Fleming, Cogenia Partners, LLC
James Marousek, Booz Allen Hamilton, Inc.
Mahsa Ettefagh, Booz Allen Hamilton, Inc.
Purser Sturgeon, Southwest Research Institute
Diego Lodato, Booz Allen Hamilton, Inc.
James Goldsmith, Booz Allen Hamilton, Inc.
Recent developments in wireless communication technologies led to the evolution of connectivity between vehicles. Maintaining connectivity between vehicles increases the awareness of the vehicle of nearby vehicles, which can be used in safety applications. Identification of malicious misbehaving vehicles play an important role in road safety. This research establishes the minimum detectable error (MDE) boundary for relative position between the observer and status vehicles (SV) using vehicle sensor and GPS error profile from field tests and established minimum standards. The results demonstrated that the MDE increases in the lateral direction (side-to-side) with the increase in relative distance between the observer and SVs while remaining the same in the longitudinal direction (front-to-back). This research effort explores the use of Sensor-Based Misbehavior Detection (SBMD) with current specifications and the defined MDE boundary for implementation in the Intersection Movement Assist (IMA) safety application to rectify false positive and false negative hazard messages propagated by a malicious misbehaving vehicle. The simulation approach used in this research effort quantifies the total number of false positive/negative hazard detection received by a third-party vehicle (TPV) using the IMA safety application and assesses the capability of the observer vehicle (OV) equipped with SBMD to rectify the false positive/negative hazard detection. In cases where there is no hazard, SBMD produces an 83% to 90% improvement in the reduction of false positive hazard detection. In the cases with hazard scenario, where the SV is in the not-safe-to-cross zone, SBMD produced an 80% to 99% improvement in application performance.
Microscopic Right-Of-Way Trading Mechanism for Cooperative Decision-Making: Theories and Preliminary Results
Zhanbo Sun (email@example.com), Southwest Jiaotong UniversityShow Abstract
Ziye Qin, Southwest Jiaotong University
Rui Ma, University of Alabama, Huntsville
Ziyan Gao, Southwest Jiaotong University
In this paper, a Microscopic Right-Of-Way Trading Mechanism (Micro-ROWTM) is developed to encourage cooperative behavior in mixed traffic (i.e., traffic mixed with cooperative vehicles and non-cooperative vehicles). Micro-ROWTM includes the following functions: (i) detection of traffic conflicts and system-improving opportunities; (ii) right-of-way trading proposal; and (iii) automatic negotiation and trade settlement. In the experiments, the proposed mechanism is tested using an illustrative ramp-merging example, in which each right-of-way trading proposal is offered to a pair of users comprised of a mainline follower and a ramp vehicle. Upon the acceptance of an offer, the mainline follower agrees to behave cooperatively (yield) to the ramp vehicle and some monetary compensation will be paid by the ramp vehicle. To help design the trading mechanism, four important definitions are introduced and discussed, including: (i) individual rationality; (ii) system-efficiency; (iii) incentive compatibility, and (iv) envy-minimization. Three different trading rules, namely, equal allocation, dynamic negotiation, and double auction are proposed and tested. Using the proposed Micro-ROWTM, the ramp vehicle will save 4.69% travel time and 34.19% fuel consumption, and the receiver will eventually get a positive benefit by compensation. Specially, the system cost can get reduced by 0.42 RMB per trade.
Empirical Analysis of Longitudinal and Lateral Vehicle Dynamics of Automated Vehicles
Julian Staiger, Munich University of Applied SciencesShow Abstract
Simeon Calvert (firstname.lastname@example.org), Delft University of Technology
The rise in availability of low-level automated vehicle systems has increased the opportunities for on-road testing in recent years. While many insights can be gained from carrying out a single pilot study, comparing and analysing the results of multiple pilots offers an increased opportunity to learn about systems and on-road performance. This paper is therefore intended to provide new insights into the automated lateral and longitudinal vehicle movement performed by (C)ACC and LCA systems based on five different on-road pilots. For longitudinal movement, a distinct difference was found between different ACC and CACC systems, where CACC systems demonstrated a more stable and homogeneous behaviour. Time-headways in acceleration phases were found to be higher compared to deceleration phases, whereby the variations are not equal. Uneven system behaviour is also visible in the evaluation of transfer functions that map the acceleration and braking dynamics of (C)ACC vehicles. Especially when accelerating, the CACC systems were found to perform well, preventing overshoots in velocity. In braking manoeuvres, the system performance of ACC and CACC systems was similar. In the case of automated lateral vehicle movement, carried out by an LCA system, the vehicle orients itself particularly to the left lane marking when driving straight ahead, while in curves it orients itself to the respective left and right lane markings. A final contribution of this paper is the presentation of a control orientated description of ACC and LCA systems that allows mapping of the considered results to be performed against the system’s general design.
Exploring the Benefits of Conversing with a Digital Voice Assistant during Automated Driving: A Parametric Duration Model of Takeover Time
Kirti Mahajan, Indian Institute of Technology, BombayShow Abstract
David Large, University of Nottingham
Gary Burnett, University of Nottingham
Nagendra Velaga (email@example.com), Indian Institute of Technology, Bombay
The current study investigated the role of an in-vehicle digital voice-assistant (VA) in conditionally automated vehicles, offering discourse relating specifically to contextual factors, such as the traffic situation and road environment. The study involved twenty-four participants, each taking two drives: with VA and without VA, in a driving simulator. Participants were required to takeover vehicle control following the issuance of a takeover request (TOR) near the end of each drive. A parametric duration model was adopted to find the key factors determining takeover time (TOT). Paired comparisons showed higher alertness and higher active workload (mean NASA-TLX rating) during automation when accompanied by the VA. Paired t-test comparison of gaze behavior prior to takeover showed significantly higher instances of checking traffic signs, roadside objects, and the roadway during the drive with VA, indicating higher situation awareness. The parametric model indicated that the VA increased the likelihood of making a timely takeover by 31%. There was also some evidence that demographic factors influenced the TOT of drivers. Male drivers likely to resume control 1.72 times earlier than female drivers. The study findings highlight the benefits of adopting a futuristic in-car voice assistant to keep the drivers alert and aware about the recent traffic environment in partially AVs.
Comfortable and Energy Efficient Velocity Control on Rough Pavements using Deep Reinforcement Learning
Jing Chen, Tongji UniversityShow Abstract
Cong Zhao, Tongji University
Yuchuan Du (firstname.lastname@example.org), Tongji University
Heavy traffic demand and high pavement maintenance burden have resulted in poor road quality and not timely improvement. These rough pavements always cause high vehicle vibration and further lead to uncomfortable sensation, high fuel consumption or even traffic crashes. Fully instrumented autonomous vehicles (AVs) make it possible to accurately detect road profiles and share the aggregated data to other connected AVs via cooperative vehicle-infrastructure system (CVIS). This paper presents an intelligent velocity control approach for AVs based on the prior knowledge of road profiles, which takes into account passenger comfort and energy efficient. The annoyance rate and jerk are used to represent discomfort caused by vehicle vibration and longitudinal acceleration, and vehicle-specific power (VSP) represents energy efficiency. Then an intelligent agent is created based on deep reinforcement learning (DRL) to learn from driving experience and integrate various influence factors to benefit vehicle driving performance and efficiency on rough pavements. The simulation environment is built by integrating Matlab and Carsim for high fidelity vehicle dynamics and suspension model, and utilizing real road profile data in Shanghai, China. The results show that the proposed velocity control approach reduces annoyance rate, jerk and VSP by 61%, 47% and 7%, respectively, which can be deployed on Robotaxi to provide high quality autonomous driving service.
COLLISION AVOIDANCE FRAMEWORK FOR AUTONOMOUS VEHICLES UNDER CRASH IMMINENT SITUATIONS
Runjia Du, Purdue UniversityShow Abstract
Sikai Chen (email@example.com), Purdue University
Yujie Li, Purdue University
Paul (Young Joun) Ha, Purdue University
Jiqian Dong, Purdue University
Samuel Labi, Purdue University
It has been widely postulated that autonomous vehicles (AVs) will profoundly increase the safety of transportation systems because automation minimizes human participation (and therefore, driver error) in the driving task. Naturally, such benefits will be fully manifest only when AV market penetration 100%; however, this is not expected to happen in the near future. The transition from a system of exclusively human-driven vehicles (HDVs) to one of exclusively AVs is expected to be lengthy. For this reason, the transition period will be characterized by a mix of HDVs and AVs in the traffic stream. Such heterogeneity causes unsafe traffic operations maneuvers due particularly to the errant nature of human driving. One of such maneuvers is high-velocity lane changing. The lane-changing process is more disruptive than other maneuvers such as car-following, and its safety consequences can be severe where the driver is inattentive or misjudges the ambient traffic conditions. This paper introduces a Model Predictive Control framework for AVs to avoid rear-end and side-impact collisions in mixed traffic streams that have aggressive lane-changing HDVs. The crash-avoidance framework utilizes V2V connectivity between HDVs and AVs, thereby allowing sharing of real-time information. The framework is tested under different traffic conditions in terms of the vehicle bumper-to-bumper distances and relative velocities. The results demonstrate the remarkable efficacy of the framework in controlling the AV movements to avoid crashes in the given traffic operations context. The success rate, which averaged at 90% for the entire process, even reached 100% where the relative velocity was low.
DISCLAIMER: All information shared in the TRB Annual Meeting Online Program is subject to change without notice. Changes, if necessary, will be updated in the Online Program and this page is the final authority on schedule information.