Stabilizing Mixed Traffic Flow using Connected Autonomous Vehicles: Consideration of Perception-reaction Time Delay and Driver Behavioral Heterogeneity of Human-driven Vehicles
Yujie Li, Purdue UniversityShow Abstract
Sikai Chen (firstname.lastname@example.org), Purdue University
Paul (Young Joun) Ha, Purdue University
Jiqian Dong, Purdue University
Runjia Du, Purdue University
Aaron Steinfeld, Carnegie Mellon University
Samuel Labi, Purdue University
The erratic nature of human driving tends to trigger undesired waves that amplify from the errant vehicle to vehicles upstream. Known as phantom jams, this phenomenon has been identified in the literature as one of the main causes of traffic congestion. This paper is based on the premise that vehicle automation and connectivity can help mitigate such jams. In the paper, we design a controller for use in a connected and autonomous vehicle (CAV) to stabilize the flow of human-driven vehicles (HDVs) that are upstream of the CAV, and consequently to lower collision risk in the upstream traffic environment. In modeling the HDV dynamics in the mixed traffic stream, we duly consider HDV driver heterogeneity and the time delays associated with their perception reaction time. We can find that the maximum number of HDVs that a CAV can stabilize, is lower when human drivers potential time delay and heterogeneity are considered, compared to the scenario where such are not considered. This result suggests that in reality, heterogeneity and time delay in HDV behavior impair the CAVs capability to stabilize traffic. Therefore, in designing CAV controllers for traffic stabilization, it is essential to consider such uncertainty-related conditions. We also show that the designed controller can significantly improve the stability of the mixed traffic stream and the safety of both CAVs and HDVs in the stream. The results are essential for effective real-world deployment of CAV controllers in mixed traffic environments.
Traffic Congestion and Shockwave Damping through Advanced Driver Assistance System (ADAS) Longitudinal Vehicle Control
Yizhou Wang (email@example.com), Rutgers UniversityShow Abstract
Peter Jin, Rutgers University
The Advanced Driver Assistance System (ADAS) has the potentials of making human driving safer, more efficient, comfortable, and environmental-friendly. Longitudinal Autonomous Vehicle (AV) control models such as Adaptive Cruise Control (ACC) and Cooperative Adaptive Cruise Control (CACC) models have been intensively studied over the last decade. Traffic shockwaves can be generated naturally by human drivers or roadway bottlenecks, such as merging sections. In some cases, if a small perturbation occurs downstream, the traffic shockwave speed could instantly propagate upstream and even lead to congestions with aggressive driving behaviors or poorly designed vehicle control models. The effect of congestion shockwaves can be further accelerated by autonomous vehicle platoons due to the instant response of those systems to preceding stimuli. This can amplify traffic instability and deteriorate the traffic conditions for the overall traffic flow and the AV platoons. This study develops ADAS longitudinal control models for AV to dampen the traffic shockwave propagation speed and mitigate the congestion. The rolling horizon control model, also called Model Predictive Control (MPC), is implemented to balance the trade-offs among multiple objectives, including safety, efficiency, and driving comfort. Vehicles models are designed separately depending on their capability of wireless communication, i.e., for ACC and CACC vehicles. The proposed models are implemented in ring-road experiments through Matlab to evaluate its stability, shockwave damping and congestion capability. The experiment results study indicates promising results of the proposed models in reducing shockwave propagation speed and mitigating traffic congestion compared with existing control methods. Keywords: Advance Driver Assistance System (ADAS), Adaptive Cruise Control (ACC), Cooperative Adaptive Cruise Control (CACC), Model Predictive Control (MPC)
Fully Distributed Optimization based CAV Platooning Control under Linear Vehicle Dynamics
Jinglai Shen (firstname.lastname@example.org), University of Maryland, Baltimore CountyShow Abstract
Eswar Kumar Hathibelagal Kammara, University of Maryland, Baltimore County
Lili Du, University of Florida
There is a surging interest in CAV platooning, supported by advanced sensing, V2V, V2I, and portable computing technologies. Various distributed optimization or control schemes have been developed for CAV platooning. However, the existing distributed schemes require either centralized data processing or centralized computation in at least one step of their schemes, referred to as partially distributed schemes. In this paper, we develop fully distributed optimization based CAV platooning control under the linear vehicle dynamics via the model predictive control approach with a general prediction horizon. The distributed schemes developed in this paper do not require centralized data processing or centralized computation through the entire schemes. To develop these schemes, we propose a new formulation of an objective function and a decomposition method that decomposes a coupled central objective function into the sum of several locally coupled convex functions whose coupling satisfies the network topology constraint. We then exploit the formulation of locally coupled optimization and operator splitting methods to develop fully distributed schemes. Control design and stability analysis is carried out to achieve desired traffic transient performance and asymptotic stability. Numerical tests demonstrate the effectiveness of the proposed fully distributed schemes and CAV platooning control.
Efficient Connected and Automated Driving System with Multi-agent graph Reinforcement Learning
Tianyu Shi, McGill UniversityShow Abstract
Jiawei Wang, McGill University
Yuankai Wu, McGill University
Luis Miranda-Moreno, McGill University
Lijun Sun, McGill University
Connected and automated vehicles (CAVs) have attracted more and more attention recently. The fast actuation time allows them to promote the efficiency and safety of transportation system. However, before having a fully autonomous transportation system, we will stay in a mixed stage where only a proportion of vehicles equipped with automation while others remain human driving. Instead of focusing on learning reliable behavior for ego automated vehicles, our research paid considerable attention on improving the outcomes of the whole transportation system by allowing each automated vehicle to cooperate with each other and regulate human-driven traffic flow. One state-of-the-art method is using reinforcement learning to learn intelligent decision making policy. However, it remains unclear whether reinforcement learning can help improve system performance with both CAVs and human driving vehicles. In this paper, we model the CAV cooperation problem in a multi-agent setting using shared policy, which achieves better system performance than non-shared policy in a single-agent setting. Furthermore, we find that the utilization of attention mechanisms on interaction features can better capture the interplay between each agent and thus boost agent cooperation. To the best of our knowledge, this work is among the first on system-level multi-agent cooperative driving using graph information sharing. We conduct extensive experiments in car-following and unsignalized intersection settings. The results demonstrate that CAVs controlled by our method can achieve the best performance against several state-of-the-art baselines.
A Framework to Determine Road Networks’ Platoonability
Egemen Okte (email@example.com), University of Illinois, Urbana-ChampaignShow Abstract
Imad Al-Qadi, University of Illinois, Urbana-Champaign
Truck platooning has several benefits over traditional truck mobility. Platooning improves safety and reduces fuel consumption up to 15%, depending on platoon configuration. Although platooning benefits are quantifiable, platooning routes are neither fully identified nor static. For efficient platooning, truck platoons need to travel at a constant high speed for extended distances. In addition, platoon integrity should be preserved from interfering vehicles that may compromise the robustness of the operation. This study presents Platoonability Level Analysis for Networks (PLAN) to determine platoonable routes based on roadway volume/capacity and number of highway exit and entry conflicts. Based on PLAN, each roadway section in a network is assigned a level of platoonability, ranging from one to five — with one being the most platoonable. PLAN was used to analyze the highway network in Illinois. It was found that 89% of the network is platoonable under average traffic conditions.
Handling Unconnected Vehicles in Cooperative Automated Driving: Adaptive Model Predictive Control Approach
Zheng Chen (firstname.lastname@example.org), University of VirginiaShow Abstract
B. Brian Park, University of Virginia
Ahmed Sakr, Toyota Motor North America
To fully harvest the benefits of cooperative automated driving in the mixed traffic consisting of both Connected and Automated Vehicles (CAVs) and non-CAVs, the usability of the Cooperative Adaptive Cruise Control (CACC) needs to be extended with more advanced control strategies. This study proposes a Cooperative Adaptive Cruise Control with Unconnected vehicle (CACCu) system, featured by Adaptive Model Predictive Control (A-MPC) approach. The proposed A-MPC CACCu enables a CAV to closely and safely follow an unconnected preceding vehicle, by making use of the information from a further preceding vehicle with connectivity. A prediction model is utilized to predict how the motion of the nearest preceding vehicle will be affected by that of the further preceding vehicle. After being initialized with a priori knowledge on the car-following behavior of a typical human driver, this prediction model is updated over time, based on the actually observed causality between the motions of the further preceding vehicle and the nearest preceding vehicle. In addition to the adaptive prediction model, a state observer is designed to make necessary corrections to the predicted acceleration of nearest preceding vehicle, by taking into account the radar measurements of spacing and relative speed. Lastly, the control command is decided by optimizing the predicted states of the ego vehicle in a rolling time horizon. The performance evaluation results using real traffic data show that the A-MPC CACCu could be robustly implemented in complicated traffic situations, and largely outperform ACC, human driving, and linear time-invariant design of CACCu.
Review of Learning-based Longitudinal Motion Planning for Autonomous Vehicles: Implications on Traffic Congestion
Hao Zhou, Georgia Institute of Technology (Georgia Tech)Show Abstract
Jorge Laval, Georgia Institute of Technology (Georgia Tech)
Anye Zhou, Georgia Institute of Technology (Georgia Tech)
Yu Wang, Georgia Institute of Technology (Georgia Tech)
Wenchao Wu, Georgia Institute of Technology (Georgia Tech)
Zhu Qing, Georgia Institute of Technology (Georgia Tech)
Srinivas Peeta, Georgia Institute of Technology (Georgia Tech)
Self-driving technology companies and the research community are accelerating their pace to use machine learning longitudinal motion planning (mMp) for AVs. This paper reviews the current state of the art in mMp, with exclusive focus on its impact on traffic congestion. The availability of congestion scenarios in current datasets are identified. The required features for training mMP are summarized. For learning methods, we surveyed the major methods in both imitation learning and non-imitation learning. The emerging technologies adopted by some leading AV giants, e.g. Tesla, Waymo, Comma.ai, are also highlighted. We find that: i) the AV industry has been mostly focusing on the long tail problem related to safety and overlooked the impact on traffic congestion, ii) current public self-driving datasets have not included enough congestion scenarios, and mostly lack the necessary input features/output labels to train mMp, and iii) albeit RL method can integrate congestion mitigation into the learning goal, the major mMp method adopted by industry is still behavior cloning (BC), whose capability to learn a congestion-mitigating mMp remains to be seen. Based on the review, the study identified the research gaps in current mMP development. Some suggestions towards congestion mitigation for future mMP studies are proposed: i) enrich data collection to facilitate congestion learning, ii) incorporate non-imitation learning methods to combine efficiency into a safety-oriented technical route, and iii) integrate domain knowledge from traditional CF theory to improve the string stability of mMP.
Exploring the effects of connected and automated vehicles at fixed and actuated signalized intersections with different market penetration rates
Li Song, University of North Carolina, CharlotteShow Abstract
Wei Fan (email@example.com), University of North Carolina, Charlotte
Pengfei Liu, University of North Carolina, Charlotte
In this paper, three widely used control systems for intelligent vehicles, i.e. the Intelligent Driving Model (IDM) for autonomous vehicles (AVs), Adaptive Cruise Control (ACC) for AVs, and Cooperative Adaptive Cruise Control (CACC) for connected and automated vehicles ( CAV s), are examined. The effects of different market penetration rates (MPRs) of intelligent vehicles in the mixed flow with HDVs at three signalized intersections (i.e. fixed signal, gap-based actuated signal, and delay-based actuated signal-controlled intersections) are further investigated under different traffic demands. The simulation results indicate that: 1) CAVs with the CACC system could decrease 49% and 96% of the average delay under low and high demand scenarios, respectively; 2) CAVs with the CACC system outperforms AVs with either ACC or IDM systems in all scenarios. CAVs with the CACC system could significantly decrease the average delay with a 20% MPRs, while significant drops could only be observed after 60% and 80% MPRs for AVs with the ACC/IDM system, respectively; 3) Gap-based and delay-based actuated signal control schemes are preferred under medium traffic demand, and CACC/ACC systems could significantly improve the performance of actuated signal-controlled intersections under high traffic demand; 4) CAVs with the CACC system could mitigate the negative effect of instable interactions with HDVs in the mixed traffic flow which are observed in the results of AVs with the ACC/IDM system. The result provides fundamental guidance for both transportation engineering researchers and practitioners to better design, plan and operate future transportation systems.
Connected and Automated Driving: Expectations and Concerns of European Citizens
Monica Grosso, European Commission Joint Research CentreShow Abstract
Amandine Duboz, European Commission Joint Research Centre
María Alonso Raposo, European Commission Joint Research Centre
Jette Krause, European Commission Joint Research Centre
Andromachi Mourtzouchou, European Commission Joint Research Centre
Fabio Luis Marques dos Santos, European Commission Joint Research Centre
Biagio Ciuffo, European Commission Joint Research Center
This paper analyses European citizens’ views on connected and automated vehicles and sheds light on how comfortable they would feel with them as part of their daily lives. The results presented are based on the 2019 Eurobarometer on “Expectations and concerns of connected and automated driving” where 27,565 European citizens were interviewed. The aim of the survey was to measure public awareness and attitudes towards connected and automated driving considering citizens’ role in defining the European strategy to improve road transport in terms of safety and efficiency. The paper, moreover, focuses on attitudes, concerns and expectations of specific user groups, in particular mobility impaired citizens and those pursuing more sustainable mobility options. The paper shows that European citizens are not yet ready to transition to these vehicles, with the majority of respondents not feeling comfortable with their presence on the roads, and a minority being willing to purchase them. In spite of the potential accessibility increase linked to automated vehicles, mobility impaired citizens seem to have a more negative attitude, while citizens striving to achieve a more environmentally friendly mobility are more positive towards these vehicles. These results clearly show the need to build trust among citizens with regard to automated vehicles. A network of Living Labs, co-creation ecosystems where people can play an active role in designing and testing new mobility concepts, can be a useful tool to build trust towards new technologies applied to the transportation system.
Investigating User Intentions of AV Use While Impaired
Diwas Thapa (firstname.lastname@example.org), University of MemphisShow Abstract
Vit Gabrhel, Transport Research Center, Czech Republic
Sabya Mishra, University of Memphis
In this study we employ Integrated Choice Latent Variable (ICLV) framework to model public’s intention 24 of using Autonomous Vehicles (AVs) while impaired. We identify five latent constructs from psychometric 25 indicators that define respondent’s perception and attitudes towards AVs which are i) perceived benefits, 26 ii) perceived risks, iii) enjoy driving, iv) wheels public transport attitude, and v) rails public transport 27 attitude. Survey data collected from 1,065 Czech residents between 2017 and 2018 is used in the study. Our 28 findings indicate that user intentions are primarily defined by attitudes rather than individual’s socio-29 demographic attributes. We found that despite acknowledgment of AVs’ potential benefits people are 30 unwilling to use AV while impaired which can largely be attributed to lack of trust in the technology.
LANE-CHANGING INTENTION IDENTIFICATION ON HIGHWAY
Zhao Yang, Chang'an UniversityShow Abstract
Hong Chen, Chang'an University
Yuwen Liu, Chang'an University
Accurate and efficient lane change intention recognition is a prerequisite for safe driving warning, which can help eliminate the interference of vehicle lateral swing and reduce the false alarm rate of the warning system. Based on the Support Vector Machine (SVM), this paper identified the driver's lane change intention through analyzing the driving characteristics and laws before lane changing. First, 13 feature vectors that can represent lane change intentions were selected, and the dimensionality reduction of each feature is conducted by the Principal Component Analysis method. At the same time, the Grid Search Method was used to find the optimal parameters. After model training, the Area under Curve (AUC) of Receiver Operating Characteristic (ROC) curve was applied to verify the model accuracy specifically. Then, the paper analyzed the phenomenon of "intention cancellation" in lane changing process, explaining the mis-classification results of partial lane keeping as lane changing. What grasped the period when lane changing intention occurs and improved precisely the performance of model recognition. Finally, vehicle trajectory data collected from I-80 expressway of NGSIM was used to verify the accuracy of this model. The results show that the model has a good recognition function and can provide some theoretical support for the development of the vehicle lane change model and vehicle early warning system on the expressway in China.Accurate and efficient lane change intention recognition is a prerequisite for safe driving warning, which can help eliminate the interference of vehicle lateral swing and reduce the false alarm rate of the warning system. Based on the Support Vector Machine (SVM), this paper identified the driver's lane change intention through analyzing the driving characteristics and laws before lane changing. First, 13 feature vectors that can represent lane change intentions were selected, and the dimensionality reduction of each feature is conducted by the Principal Component Analysis method. At the same time, the Grid Search Method was used to find the optimal parameters. After model training, the Area under Curve (AUC) of Receiver Operating Characteristic (ROC) curve was applied to verify the model accuracy specifically. Then, the paper analyzed the phenomenon of "intention cancellation" in lane changing process, explaining the mis-classification results of partial lane keeping as lane changing. What grasped the period when lane changing intention occurs and improved precisely the performance of model recognition. Finally, vehicle trajectory data collected from I-80 expressway of NGSIM was used to verify the accuracy of this model. The results show that the model has a good recognition function and can provide some theoretical support for the development of the vehicle lane change model and vehicle early warning system on the expressway in China.
The Impact of the Choice of Liability Regime on Automated Vehicles’ Driving Behavior and Network Throughput
Danqi Shen (email@example.com), Southwest Jiaotong UniversityShow Abstract
Scott Le Vine, SUNY New Paltz
Xiaobo Liu, Southwest Jiaotong University
The impacts of highly-automated vehicles have received growing attention as the technology advances towards commercialization. One of the areas of focus has been their impact on traffic flow properties. However, while their driving behavior will be constrained by the legal regime in which they will operate, the relationships between choice of legal regime and driving behavior (and their behavior’s system-level consequences on traffic flow) have yet to be established. This is a timely issue, as regulation of automated vehicles is as yet unsettled. Negligence Liability in general yields economically efficient outcomes, if perfect information is available. However, in the case of human driving there is a major breach of the perfect information assumption, and Strict Liability is typically applied, but the advent of rich, high-fidelity data streams associated with Vehicle Automation may make a shift to Negligence Liability possible. In this paper, we apply the logic of the Strict and Negligence Liability regimes as well as No-Fault, and evaluate their differential consequences for network throughput. In numerical analyses using empirical vehicle-trajectory data, we demonstrate that automated vehicles’ optimal following distances under the Negligence Liability regime may be smaller, yielding 11-12% greater throughput than Strict Liability. We also demonstrate that the proposed approach enables designers of AVs’ driving behavior to take account of marketability (i.e. consumer preferences) in managing their exposure to liability. It is hoped that exposition of these issues will support informed design and regulation of automated vehicles. The paper concludes with a brief discussion of future research needs.
Field Experiments of Commercially Available Automated Vehicles on Freeways
Pablo Chon Kan, Florida Atlantic UniversityShow Abstract
Servet Lapardhaja, University of California, Berkeley
Xingan (David) Kan, Florida Atlantic University
The first generation of autonomous vehicles are already commercially available and can maintain the desired speed and following the preceding vehicles autonomously via Adaptive Cruise Control (ACC). ACC utilizes data that are collected from the on-board sensors to enable automated car following. Field experiments demonstrate that today’s commercially available ACC vehicles provide similar headways and capacities as human driven vehicles on freeways under steady-state and free-flow conditions. However, field tests also demonstrated that the design of today’s commercially available ACC vehicles can lead to delayed response and gradual acceleration when operating on freeways with speed fluctuations such as at a bottleneck. Both the delayed response and the gradual acceleration can lead to increase in headway thus decrease in capacity at bottlenecks were there are queues and significant fluctuations in speeds. The capacity reduction can be as much as 36%.
Lane Keeping and Lane Changing of Automated Control Based on Driving Safety Domain
CHUAN SUN, Tsinghua UniversityShow Abstract
Sifa Zheng, Tsinghua University
Yulin Ma, Tsinghua University
Li Wei, Huanggang Normal University
Duanfeng Chu, Wuhan University
Junru Yang, Wuhan University
Yicheng Li, Jiangsu University
As the complex driving scenarios bring about an opportunity for application of deep learning in safe driving, artificial intelligence (AI) based on deep learning has become a heatedly discussed topic in the field of advanced driving assistance system (ADAS). This paper focuses on analysing vehicle active safety control of collision avoidance for intelligent connected vehicles (ICVs) in a real driving risk scenario, and driving risk perception is based on the ICV technology. In this way, trajectories of surrounding vehicles can be predicted and tracked in a real-time manner. In this paper, vehicle dynamics based state-space equations conforming to model predictive controllers are set up to primarily explore and identify a safety domain of active collision avoidance. Furthermore, the model predictive controller is also designed and calibrated, thereby implementing the active collision avoidance strategy for vehicles based on the model predictive control method. At last, functional testing is conducted for the proposed active collision avoidance control strategy in a designed complex traffic scenario. The research findings here can effectively improve automatic driving, intelligent transportation efficiency and road traffic safety.
NDP-TP3: An Intersection Management Framework for Signal Optimization and Trajectory Control in Mixed Connected Automated Traffic
Yi Guo, University of CincinnatiShow Abstract
Jiaqi Ma (firstname.lastname@example.org), University of California, Los Angeles
The emerging connected and automated vehicle (CAV) technologies offer new opportunities for urban signalized intersection management. Through wireless communication and advanced sensing capabilities, CAVs can detect the surrounding traffic environment and share real-time vehicular information with others or infrastructure while individual trajectories of CAVs can be precisely controlled. This paper proposes a real-time adaptive intersection management framework for signal optimization and trajectory control. The proposed framework aims at improving traffic efficiency in terms of throughput and delay in mixed connected automated traffic. This framework utilizes neuro-dynamic programming to solve signal optimization problems with observed traffic states in real time. The vehicular trajectories of CAVs can be precisely controlled to maximize the utilization of the green time and reduce the start-up lost time by using a trajectory planning algorithm. Since human-driven vehicles’ vehicular information cannot be directly obtained in the mixed-autonomy environment, a vehicle status estimation function is developed by using CAV data. The CAV platooning operation is also implemented such that CAVs can pass the intersection efficiently, and more importantly, to avoid excessive centralized control of all CAVs. Simulations are performed at a signalized intersection, and results show that the proposed NDP-TP3 framework can significantly improve the throughput and reduce the average delay across different CAV market penetration rates (MPRs). The effectiveness of subset strategies of the NDP-TP3 is also investigated. The results show that the signal optimizer can effectively allocate green phases. The trajectory control can benefit the traffic system at medium to high CAV MPRs and can be adversely affected by human-driven vehicles at lower CAV MPRs.
Cooperative Perception for Estimating and Predicting Macroscopic Traffic States
Tao Li, University of California, Los AngelesShow Abstract
Jiaqi Ma (email@example.com), University of California, Los Angeles
Connected and automated vehicles (CAV) are equipped with different sensing technologies such as radar, camera, and LIDAR. They interact with other non-observable vehicles constantly in the traffic stream and provide real-time data of themselves and surrounding vehicles. With a comprehensive fusion of the multi-sensor datasets, we can enhance the observability of the traffic streams and provide more accurate estimations and predictions of the traffic states. In this paper, we proposed a cooperative perception framework based on the Cell Transmission Model (CTM) and Unscented Kalman filter. Cooperative perception is performed using the CAV data to enhance the lane-based high-definition (HD) traffic state estimation and prediction by sharing information collected from the CAV onboard sensors (i.e., radar or lidar-sensors). The results show that when the CAV market penetration rate (MPR) is 25%, the mean absolute percentage errors (MAPE) for speed estimation are between 0.01 and 0.1 and the density estimation MAPE results are between 0.1 and 0.2. When the CAV MPR is between 5%-10%, the speed estimation MAPE is between 0.05 and 0.2, while the density estimation MAPE is between 0.1 and 0.3. The results also suggest that better sensing capability with multiple observations around the CAVs will improve the performance and decrease the MAPE by about 0.05-0.2 depending on the MPR. Generally, the proposed framework provides efficient and accurate estimations and predictions of macroscopic local traffic states , providing the dynamic traffic environment conditions as inputs to enable fine control and management of the connected and automated traffic systems.
Freeway Cooperative Merge Control via Centralized Multi-Agent Deep Reinforcement Learning
Xu Han, University of CincinnatiShow Abstract
Jiaqi Ma (firstname.lastname@example.org), University of California, Los Angeles
Connected and Automated Vehicles (CAVs) is a promising solution to real-world traffic problems. Naturally, CAV traffic control can be modeled as reinforcement learning (RL) systems. However, methods that efficiently coordinate multiple CAVs in a complex environment is still in the infancy. To this end, we propose a multi-agent deep RL method that controls all CAVs in a freeway on-ramp merging scenario. The proposed method uses a centralized actor-critic RL framework. To increase the estimation accuracy, it utilizes a convolutional deep neural network (CNN) that extracts speed, location, and vehicle type information. Also, to cope with the challenge of exponentially expanding search space when controlling several CAVs, the proposed method considers CAVs' joint actions when estimating the Q-value, which limits the search space and stabilizes the training. The proposed method also adopts the N-step temporal difference learning (TD-N learning) for loss function calculation. The TD-N learning optimizes the critic's DQN network based on the future N-step Q-values, which accounts for the delayed impact on traffic along the upstream direction. We trained and evaluated our model by deploying it in the simulation with different scenarios of traffic congestion and CAV market penetration rates. Results show the proposed method can reduce the network delay by 30%-50% with mainline CAV cooperation. Furthermore, the model with CAV cooperation across all lanes shows am additional 27% delay reduction versus the mainline cooperation only, though both cooperation scenarios result in significant traffic improvement as compared to the non-CAV scenario.
A model-based validation framework for safety evaluation of autonomous systems
Linda Capito, Ohio State UniversityShow Abstract
Umit Ozguner, Oztech Inc
The deployment of autonomous vehicles in the foreseeable future poses a huge challenge in the validation and verification of the safety operation of these systems. Unlike previous generations systems, which operation and safety could be defined by standards such as the ISO 26262, novel autonomous systems become increasingly complex to analyze due to the exploding number of possible system configurations. To attack this problem, we propose to study the safety case for challenging scenarios by proposing a methodology for model based validation. This method combines a functional hierarchical decomposition approach to get a better understanding of the autonomous system, and a dynamic probabilistic risk assessment algorithm called backtracking process algorithm to perform risk analysis. This method allows to provide a good coverage of the different system configurations that may take place for a predefined scenario. The use of this methodology is tested in a complex system that consists of two autonomous trucks that travel together on the highway. Three relevant scenarios from this system are chosen and analyzed using the proposed method. Finally, we identify risks associated with the choice of acceleration/deceleration parameters, presence of a communication link and external vehicles interfering with the truck system.
Analysis of Advanced Driver Systems for Lane Keeping Aids
Kamal Windom, Prairie View A&M UniversityShow Abstract
Sarhan Musa, Prairie View A&M University
This paper analyzes the Advanced Driver Assistance Systems (ADAS) scaling diagrams of driving steering override in the case of a user behind the wheel. To assure driver safety, the ADAS displays resistance to deter accidental deviance from the lane. In addition, we analyze the control system relationship between the ADAS, Electronic Stability Control (ESC) and the driver, and break down a possible solution to prevent lane deviance. The ESC offers a differing level of torque resistance to the user turning the wheel. Based on the lateral position of the car on the road, the ADAS can use the ESC to increase or decrease torque output. The study between the scaling factor and torque output can be analyzed through Matlab. Furthermore, the function between the scaling factor and torsion bar torque can be theoretically changed by altering variables.
Connected and Automated Vehicles On-Ramp Merging Strategy Based on Complete Information Static Game
Yukun Fang, Chang'an UniversityShow Abstract
Haigen Min, Chang'an University
Zhigang Xu, Chang'an University
Xiangmo Zhao, Chang'an University
Vehicles on-ramp merging is one of the main causes that leads to the reduction of traffic efficiency and fuel economy, and increases the risk of collision. Cooperative control for connected and automated vehicles (CAVs) has the potentials to signigicantly reduce the negative environmental impact and improve the safety and traffic efficiency. Therefore, in this paper, we focus on the scenario of CAVs on-ramp merging and propose a centralized control method. The merging sequence (MS) allocation and motion planning are the key issues affecting the traffic conditions. To deal with these two problems, we firstly propose a MS allocation method based on complete information static game, where the mixed strategy Nash equilibrium is the basis for the individual vehicle to select their strategies. Then, the on-ramp merging problem is formulated as a bi-objective optimization problem and optimal control are applied to solve the motion planning issue. To determine the proper weight parameters in the bi-objective optimization problem, varying scale grid search method is proposed to explore possible solutions in different scales. In this method, an improved quick sort algorithm is designed to search for the Pareto front, and the (approximately) unbiased Pareto solution for the bi-objective optimization problem is finally determined as the optimal solution. Availability of the proposed model is validated via simulation, and comparison with other models demonstrates the effectiveness of the proposed model in fuel economy and traffic efficiency.
Integrated Longitudinal and Lateral Decentralized Control of Connected and Automated Vehicles Cooperative Merging at On-ramps
Shoucai Jing (email@example.com), Chang'an UniversityShow Abstract
Fei Hui, Chang'an University
Xiangmo Zhao, Chang'an University
Jackeline Rios-Torres, Oak Ridge National Laboratory
Asad Khattak, University of Tennessee
Connected and automated vehicles (CAVs) can improve traffic safety and transportation network efficiency while also reducing environmental impacts. However, congestion and accidents can easily occur at merging roadways. Therefore, coordinating cooperative merging of CAVs is one of the most common and respective traffic management problems. This paper addresses the problem of integrated longitudinal and lateral cooperative merging control with practical implications for CAVs approaching on-ramps. A hierarchical and decentralized cooperative coordination framework was developed to systematically control CAVs merging. The control system of each vehicle is divided into an upper and lower level. An optimal control based algorithm considering input constraints is presented, which optimizes fuel consumption and passenger comfort. Furthermore, a decentralized unified algorithm for lower level lateral control is proposed for tracking the upper-level optimal trajectory based on nonlinear model predictive control. The driving safety field is used as the one of optimization objectives to avoid lateral collision. Efficiency of the proposed framework and algorithm were validated using simulations of near real-world scenarios in CarSim/Simulink. The proposed integrated merging control system can improve traffic efficiency and reduce fuel consumption compared to baseline with the potential for real-world application. Furthermore, the results demonstrate the potential applicability of cooperative control methods based on upper level vehicle control.
Freeway Cooperative Merge in Dense Traffic Using Decentralized Genetic Fuzzy Systems
Anoop Sathyan (firstname.lastname@example.org), University of CincinnatiShow Abstract
Jiaqi Ma, University of California, Los Angeles
Kelly Cohen, University of Cincinnati
Cooperative ADS equipped Vehicles (CAVs) can be a promising solution to real-world traffic problems. The problem of cooperative merge is of particular interest. It is very important that vehicles moving along the mainline and the merging vehicles cooperate to ensure smooth flow of traffic to avoid congestion. This paper presents the application of a reinforcement learning strategy using genetic fuzzy systems (GFS) for cooperative merge of vehicles onto a highway in high density, mixed traffic conditions. The CAVs are trained to make their own decisions, thus making this a decentralized system. The CAVs make decisions purely based on local information. The GFS module in each CAV makes recommendations on its speed and lane changes. The CAVs are trained on mixed traffic scenarios with 50% market penetration rates (MPR) and then tested on different scenarios with varying MPRs. The results show the effectiveness of increasing the CAV MPR. As the system was trained on mixed traffic, CAVs learn to work with human-driven vehicles that are modeled in the simulation environment using simpler vehicle following and lane change models. The trained GFS model has been tested on two different traffic volumes across different CAV MPRs.
Evaluating Cost Savings of Truck Caravanning
Vasileios Liatsos (email@example.com), University of MemphisShow Abstract
Dimitrios Giampouranis, University of Memphis
Mihalis Golias, University of Memphis
Sabya Mishra, University of Memphis
Razi Nalim, Indiana University
Mark Frohlich, Indiana University
Clayton Nicolas, Indiana University
In this paper we evaluate an alternative to the truck platooning concept known as truck caravanning. Truck platooning and related autonomous caravanning concepts constitutes a sustainable solution to truck entrepreneurs to increase their profits and improve their service quality. The goal of this paper is not to evaluate the technical feasibility of truck caravanning but rather propose a preliminary model that can evaluate cost savings. The rational is that unless substantial monetary savings exist to justify the initial and high capital investment (both in equipment and infrastructure) we should not be investing heavily in research and development of the technology behind caravanning. Numerical results indicate that when the size of the caravan equals or exceeds four trucks significant cost savings can be achieved.
Stated Preference Analysis of Autonomous Vehicles Among California Residents Using Probabilistic Inferences
Jimoku Salum, Florida International UniversityShow Abstract
Boniphace Kutela, Texas A&M Transportation Institute
Angela Kitali, Florida International University
Emmanuel Kidando, Cleveland State University
As technology advances, improvements in the way people travel also change. In 2017, the National Renewable Energy Laboratory (NREL) conducted a travel survey in California to understand the residents' perception of several mobility aspects. This study used the data collected by NREL to understand various factors associated with the safety perception and acquisition of autonomous vehicles among California residents. Bayesian Networks (BNs) was used to learn the probabilistic interrelationship among autonomous vehicles' aspects . The predicted probabilities for the safety concern of self-driving vehicles, purchase of vehicles with auto-drive assistance, and purchase of self-driving vehicles were determined after learning the BN structure and parameters from the data. The study found that there is a strong relationship between the acquisition of autonomous vehicles and vehicles with auto-drive assistance. The BN model predicted that residents who are interested in purchasing vehicles with auto-drive assistance are about 95% likely also to purchase self-driving vehicles. Moreover, the ridesharing, number of vehicles in the household, housing type, and Pug-in Vehicle (PEV) ownership are among the factors that play a great role in the acquisition and safety perception of autonomous vehicles. Residents who are currently participating in ridesharing and living in the apartments are more likely to purchase the vehicles with auto-drive assistance. Residents who either own the PEV or have three or more vehicles are more likely to have safety concerns with self-driving vehicles. Additionally, residents who do not have safety concerns with self-driving vehicles are about 45% likely to purchase them.
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.