Conference PaperPDF Available

Deep Learning-based Human-Driven Vehicle Trajectory Prediction and its Application for Platoon Control of Connected and Autonomous Vehicles

Authors:
Lei Lin
Lei Lin
Lei Lin
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Siyuan Gong at Chang'an University
Siyuan Gong
Tao Li at Purdue University West Lafayette
Tao Li
Download full-text PDF
Read full-text
Download citation

Abstract and Figures

The advent of connected and autonomous vehicles (CAVs) will change driving behavior and the travel environment, providing opportunities for safer, smoother, and smarter road transportation. During the transition from the current human-driven vehicles (HDVs) to a fully CAV traffic environment, the road traffic will consist of a “mixed” traffic flow of HDVs and CAVs. Equipped with multiple sensors and vehicle-to-vehicle communications, a CAV can track the trajectories of other CAVs in its vicinity, and ideally, all CAVs in communication range. Such CAV trajectory data can be leveraged with advances in computing and machine learning algorithms to potentially predict trajectory data of HDVs, such as acceleration and speed. Based on these predictions, CAVs can react accordingly to avoid or mitigate traffic flow oscillations and accidents. In this study, we seek to predict a HDV’s trajectory based on two types of deep learning models: (i) the Long Short-Term Memory (LSTM) model, which is efficient for temporal trajectory prediction because its neural network architecture is designed to utilize inputs from previous time steps and (ii) a model that combines two deep learning architectures, the LSTM and the Convolutional Neural Network (CNN) model. CNN provides the capability to feed more information into the LSTM using images. The images are converted from time series trajectory data, road geometry data and relative vehicle positions. As a case study, these two deep learning models will be used to predict the leading HDV trajectory for platoon control of CAVs; that is, in this scenario a HDV is the leading vehicle for a group of CAVs to platoon. To dampen traffic flow oscillations, model predictive control (MPC), which can fully leverage deep learning-based predictions, is implemented to generate the control law for each CAV in the platoon. The performance of the MPC platoon control with trajectory predictions using the two deep learning models will be compared to that of CAV platoon control based on a traditional trajectory prediction model.
Figures - uploaded by Tao Li
Author content
All figure content in this area was uploaded by Tao Li
Content may be subject to copyright.
Content uploaded by Tao Li
Author content
All content in this area was uploaded by Tao Li on Sep 22, 2018
Content may be subject to copyright.
1/30
Deep Learning-based Human-Driven Vehicle Trajectory
Prediction and its Application for Platoon Control of
Connected and Autonomous Vehicles
Lei Lin1Siyuan Gong1Tao Li2
1NEXTRANS Center
Purdue University
2Department of Computer Science
Purdue University
Automated Vehicles Symposium, 2018
1 / 30
2/30
Outline
1Problem Description
2Motivation
Mixed Traffic Flow
Data-driven Model
3Data Set and Preprocessing
4Model Building
5Results
LSTM Prediction Performances
CAV Platoon Control Performances
2 / 30
3/30
1Problem Description
2Motivation
3Data Set and Preprocessing
4Model Building
5Results
3 / 30
4/30
Problem Description
Predict the leading Human-driven Vehicle’s (HDV) trajectory for
platoon control of CAVs.
4 / 30
5/30
1Problem Description
2Motivation
Mixed Traffic Flow
Data-driven Model
3Data Set and Preprocessing
4Model Building
5Results
5 / 30
6/30
Motivation-Mixed Traffic Flow
Long transition period to a fully Connected and Autonomous Vehicle
(CAV) environment.
All new vehicles will have the connectivity function until 2025 (GSMA,
2013).
75% of vehicles will be autonomous until 2040 (IEEE, 2012).
6 / 30
7/30
Motivation-Data-driven Model
Most vehicle trajectory prediction models are parsimonious.
Provide analytical conclusions and fast simulation speed.
Limited model flexibility and accuracy.
For example, car following models in traffic simulation studies rely on
artificial parameters in empirical equations.
Trajectory data from CAVs in the communication range can help to
predict the HDV’s trajectory.
7 / 30
8/30
1Problem Description
2Motivation
3Data Set and Preprocessing
4Model Building
5Results
8 / 30
9/30
Data Set and and Preprocessing
Next Generation Simulation
(NGSIM) Data Set:
500 meters long, 4:00
PM-4:15 PM, 04/13/2005,
I-80.
Lateral and longitudinal
locations, VehID, Time Step,
Velocity, Acceleration,
Preceding VehID, LaneID by
10 Hz.
Reconstructed data were
downloaded (Punzo et al.,
2011).
1
1https://www.fhwa.dot.gov/publications/research/operations/06137/index.cfm
9 / 30
10/30
Data Set and and Preprocessing
Data Preprocessing Steps:
CAV Market Penetration Rate: 50%.
For lane 2, 3, and 4, find the Ego CAV and HDV pairs, and for each
pair, find 3 CAVs in the front 300 meters from left, current, and right
lanes separately.
Record the time step, lane location, speed, acceleration rate, vehicle
length for the Ego CAV, the HDV and 9 CAVs.
10 / 30
11/30
1Problem Description
2Motivation
3Data Set and Preprocessing
4Model Building
5Results
11 / 30
12/30
Model Building
2
The Long Short-term Memory Model (LSTM).
Designed for Sequencial Data.
Consists of LSTM Cells, each cell has an input gate, forget gate and
output gate.
Automatically memorizing relevant spatial and temporal features
extracted from previous vehicle states.
2http://colah.github.io/posts/2015-08-Understanding-LSTMs/
12 / 30
13/30
Model Building
Experimental Environment:
GTX 1080, Tensorflow.
LSTM Architecture:
Learning rate: 0.001.
LSTM cell number:100.
Mini-batch size: 100.
Hidden units in each cell: 400.
Early stopping threshold: 100.
Data Set:
Training dataset size:
29376 X 7750.
Validation dataset size:
9792 X 7750.
Testing dataset size:
9792 X 7750.
13 / 30
14/30
1Problem Description
2Motivation
3Data Set and Preprocessing
4Model Building
5Results
LSTM Prediction Performances
CAV Platoon Control Performances
14 / 30
15/30
Results-LSTM Prediction Performances
Table 1: RMSE of Testing Data Sest.
Number of CAVs Prediction Steps
10 50
3 CAVs 0.67 0.72
9 CAVs 0.71 0.72
Table 2: Normalized RMSE of Testing Data Sest. (1-RMSE/MEAN)
Number of CAVs Prediction Steps
10 50
3 CAVs 95.40% 95.11%
9 CAVs 95.19% 95.01%
15 / 30
16/30
Results-LSTM Prediction Performances
RMSE by HDV ID (268 HDVs)
10 Prediction Steps with 3 CAVs 50 Prediction Steps with 3 CAVs
16 / 30
17/30
Results-LSTM Prediction Performances
The Minimum and Maximum RMSEs for one Round of Prediction (10
Prediction Steps with 3 CAVs)
17 / 30
18/30
Results-LSTM Prediction Performances
The Minimum and Maximum RMSEs for one Round of Prediction (50
Prediction Steps with 3 CAVs)
18 / 30
19/30
Results-CAV Platoon Control Performances
Select two representative HDVs, assume each is followed by a platoon of 9
CAVs, apply the P-step Model Predictive Control (MPC) (Gong and Du,
2018):
Table 3: RMSEs of Two HDVs.
HDV Prediction Steps
10 50
2973 0.56 0.64
3307 0.94 0.94
19 / 30
20/30
Results-CAV Platoon Control Performances
Impact of MPC on the speed of the CAV platoon following HDV 2973:
10-step predictions
50-step predictions 20 / 30
21/30
Results-CAV Platoon Control Performances
Impact on the acceleration rate of the CAV platoon following HDV 2973:
10-step predictions
50-step predictions 21 / 30
22/30
Results-CAV Platoon Control Performances
Impact on the spacing of the CAV platoon following HDV 2973:
10-step predictions
50-step predictions 22 / 30
23/30
Results-CAV Platoon Control Performances
Select two representative HDVs, assume each is followed by a platoon of 9
CAVs, apply the P-step Model Predictive Control (MPC) (Gong and Du,
2018):
Table 4: RMSEs of Two HDVs.
HDV Prediction Steps
10 50
2973 0.56 0.64
3307 0.94 0.94
23 / 30
24/30
Results-CAV Platoon Control Performances
Impact of MPC on the speed of the CAV platoon following HDV 3307:
10-step predictions
50-step predictions 24 / 30
25/30
Results-CAV Platoon Control Performances
Impact on the acceleration rate of the CAV platoon following HDV 3307:
10-step predictions
50-step predictions 25 / 30
26/30
Results-CAV Platoon Control Performances
Impact on the spacing of the CAV platoon following HDV 3307:
10-step predictions
50-step predictions 26 / 30
27/30
Summary
Contributions
Utilized information from multiple vehicles to predict the HDV’s
acceleration rates.
Applied the state-of-the-art LSTM deep learning model for
multiple-step-ahead predictions.
Based on our predictions, MPC Control can ensure both transient
traffic smoothness and asymptotic stability of the CAV platoon.
Limitations
Assumption of ”connected”.
Assumption of ”automated”.
27 / 30
28/30
Next Steps
Further clean the dataset.
Build the CNN-LSTM model.
Test other platoon control algorithms.
Evaluate the performances under different CAV MPRs, CV
communication ranges and delay and so on.
Predict a distribution of the acceleration rate and capture the
uncertainty.
28 / 30
29/30
References
GSMA, 2013. Eevery New Car Connected by 2025 as Embedded
Mobile Mobile Technology Drives Growth of Connected Car Market.
IEEE, 2012.
https://www.ieee.org/about/news/2012/5september-2-2012.html
(accessed 6.1.18).
Punzo, V., Borzacchiello, M.T. and Ciuffo, B., 2011. On the
assessment of vehicle trajectory data accuracy and application to the
Next Generation SIMulation (NGSIM) program data. Transportation
Research Part C: Emerging Technologies, 19(6), pp.1243-1262.
Siyuan Gong and Lili Du. 2018. Cooperative Platoon Control for a
Mixed Traffic Flow Including Human Drive Vehicles and Connected
and Autonomous Vehicles. Transportation Research Part B:
Methodological. Accepted.
29 / 30
30/30
Thanks for your attentions!
Any questions?
30 / 30
... Machine learning based techniques for vehicle trajectory nowcasting, i.e., short-term trajectory forecasting [1], play a central role in future 5G/6G radio access networks (RAN) as a key enabler of road safety algorithms for autonomous vehicles [2] or of anticipatory RAN resource management strategies [3], to mention a few. Machine learning (ML) based techniques for vehicle trajectory nowcasting, i.e., short-term trajectory forecasting [1], play a central role in future 5G/6G vehicular networks as a key enabler of anticipatory radio access network (RAN) resource management strategies [3], as well as road safety algorithms for autonomous vehicles [2]. ...
... Machine learning based techniques for vehicle trajectory nowcasting, i.e., short-term trajectory forecasting [1], play a central role in future 5G/6G radio access networks (RAN) as a key enabler of road safety algorithms for autonomous vehicles [2] or of anticipatory RAN resource management strategies [3], to mention a few. Machine learning (ML) based techniques for vehicle trajectory nowcasting, i.e., short-term trajectory forecasting [1], play a central role in future 5G/6G vehicular networks as a key enabler of anticipatory radio access network (RAN) resource management strategies [3], as well as road safety algorithms for autonomous vehicles [2]. Several challenges, however, stand in the way of designing a scalable and efficient strategy for training an ML prediction model for vehicular trajectory nowcasting. ...
... The issue of short-term vehicular trajectory prediction has recently received much attention, given the growing amount of applications and use cases, both in ITS and Autonomous driving domain, and in 5G and beyond network optimization and dynamic management [45], [46]. The growing availability of a large amount of car/user floating data has generated a large body of works which apply various deep learning techniques to the problem of predicting network-wide vehicle movement patterns in urban scenarios [2], [45]- [47]. [48] applies a LSTM architecture to the problem of vehicle nowcasting in urban scenarios based on a centralized scheme. ...
A Gossip Learning Approach to Urban Trajectory Nowcasting for Anticipatory RAN Management
Article
  • Jan 2023
In future radio access networks, machine learning (ML) based strategies for short-term forecasting of vehicular trajectories will be key for anticipatory resource allocation and management at the mobile edge. However, training ML models in a centralized fashion, over data collected from a massive heterogeneous and dynamic set of devices, poses significant scalability, reliability, and efficiency challenges, which are still open to date. In this paper, we look at the specific issue of scalable and resource-efficient training of ML models in a vehicular environment. To address such a challenge, we propose a new Gossip Learning scheme, i.e., a fully distributed, collaborative training approach based on direct, opportunistic model exchanges via wireless device-to-device (D2D) communications with no centralized support. Our approach is based on constantly improving each node's own model instance through knowledge transfer among nodes, and on different strategies for estimating the potential contribution of neighboring nodes to the training process at a node. Extensive numerical assessments on a variety of measurement-based dynamic urban scenarios suggest that our schemes are able to converge rapidly and provide sufficiently accurate forecasts of vehicle position for time horizons which are typical of future 5G/6G dynamic resource allocation algorithms.
View
... ML integration with MPC, which sometimes refer as learning MPC (Hewing et al., 2020), or Data-driven MPC (Piga et al., 2019), is divided into three main categories of: (1) In offline modeling, the coefficient and structure of the ML model are developed offline using measurement data, and this model is used to develop the MPC controller. Offline learning of a model were widely used in literature including application to a variety of industries such as HVAC (Chen et al., 2020a,a), chemical process (Chen et al., 2020b,c;, bio-process (Bradford et al., 2020), wind energy robotics (Gillespie et al., 2018), and extensively in the automotive industry (Wang et al., 2006;Dahunsi et al., 2009;Lin et al., 2018;Huang et al., 2019;Xie et al., 2019;Khoshbakht Irdmousa et al., 2019;Basina et al., 2020;Arab and Yi, 2020;Iwadare et al., 2009;Janakiraman et al., 2016;Šika et al., 2008;Lu et al., 2015;Marx et al., 2012;Xie et al., 2018;Alcala et al., 2020;Da Rocha et al., 2016;Garimella et al., 2017). ...
... So various types of cells with long-term memory were introduced to tackle this problem. The most popular and well-known of these long-term memory cells is the Long Short-Term Memory cell (LSTM) (Lin et al., 2018). The LSTM cell was first developed in 1997 and then gradually improved by several researchers. ...
... Unlike RNN, the integration of LSTM and MPC has only been implemented rarely. In (Lin et al., 2018), LSTM was used to predict human-driven vehicles' trajectory in platooning control of connected and autonomous vehicles. In this study, the speed and acceleration rate of the leading vehicle were predicted using LSTM, and p-step MPC with the different p-values were analyzed with the assumption that vehicles must be connected and automated. ...
Integrating Machine Learning and Model Predictive Control for Automotive Applications: A Review and Future Directions
Article
  • Jan 2023
  • ENG APPL ARTIF INTEL
In this review paper, the integration of Machine Learning (ML) and Model Predictive Control (MPC) in Automotive Control System (ACS) applications are discussed. ACS can be divided into these three main subsystems: enhancing safety, improving comfort, and reducing fuel consumption and emissions. Due to the development of new technologies such as advancing autonomous and connected vehicles the complexity of these subsystems is increasing. The ACS is meant to encompass the vehicle dynamics, powertrain control, passenger comfort, and accessories. Since vehicle manufacturers must meet stringent performance and emission requirements, optimal control methods for ACS applications are seen as a promising technology. MPC is an optimal control method for closed-loop control applications that allows constraints to be enforced in real-time while an objective function is minimized. The application of MPC in the automotive industry has been shown in the past decade. An important challenge in the design and real-time implementation of MPC is having a accurate predictive model that also does not require excessive real-time computation. Using ML to provide an accurate model at decreased computational cost improves MPC performance of ACS and is the main focus of this paper. How MPC in ML-based ACS applications ensures stability while meeting constraint is also discussed. Method to combine MPC and ML for the ACS subsystems of vehicle dynamics and powertrain control are reviewed and an outlook on future ACS is discussed.
View
... These networks are vulnerable to attacks that can compromise the functioning of the vehicle. An attacker who gains access to an in-vehicle network can potentially control the vehicle's systems, including the brakes, steering, and acceleration [25]. There is a need for advanced intrusion detection systems and secure network architectures to mitigate the risks associated with in-vehicle network attacks [26,27]. ...
... Usage of fog server with fog anonymizer to avoid eavesdropping in vehicular ad-hoc networks (VANETs) [20] P5 Maintaining data integrity in dynamic route guidance by forged data filtering scheme [25] P6 Using swarm algorithms for routing attacks [37] P7 Detecting bandwidth and entropy to reduce denial of service attack [38] P8 Implementing noisy control signals to avoid replay attacks [39] P9 ...
A fsQCA-Based Framework for Cybersecurity of Connected and Automated Vehicles: Implications for Sustainable Development Goals
Article
Full-text available
  • Feb 2024
Connected and automated vehicles (CAV) are increasingly recognized as a critical component of intelligent transportation systems (ITS), contributing to advances in transportation safety and mobility. However, the implementation of CAV in a real-world environment comes with various threats, and cybersecurity is among the most vulnerable. As the technology becomes more advanced and complex, it is essential to develop a comprehensive cybersecurity framework that can address these concerns. This research proposes a novel framework based on complexity theory and employs the fuzzy set qualitative comparative analysis (fsQCA) technique to identify combinations of security attacks that lead to achieving cybersecurity in CAV. Compared to structural equation modelling (SEM), the fsQCA method offers the advantage of demonstrating all possible ways to achieve the outcome. The study’s findings suggest that in-vehicle networks and data storage security are the most crucial factors in ensuring the cybersecurity of CAV. The results can be useful for automotive designers in reducing the potential for attacks while developing secure networks.
View
... It has been successfully applied for natural language text compression, speech/handwriting recognition, etc. In the transportation domain, it was applied for traffic speed prediction (Ma et al., 2015), vehicle trajectory prediction (Lin et al., 2018b) and so on. More recently, the integration of the LSTM architecture with the CNN architecture has been reported to improve large-scale taxi demand predictions by modeling both spatial and temporal relationships (Ke et al., 2017;Yao et al., 2018). ...
Predicting Station-level Hourly Demands in a Large-scale Bike-sharing Network: A Graph Convolutional Neural Network Approach
Preprint
Full-text available
  • Dec 2017
This study proposes a novel Graph Convolutional Neural Network with Data-driven Graph Filter (GCNN-DDGF) model that can learn hidden heterogeneous pairwise correlations between stations to predict station-level hourly demand in a large-scale bike-sharing network. Two architectures of the GCNN-DDGF model are explored; GCNNreg-DDGF is a regular GCNN-DDGF model which contains the convolution and feedforward blocks, and GCNNrec-DDGF additionally contains a recurrent block from the Long Short-term Memory neural network architecture to capture temporal dependencies in the bike-sharing demand series. Furthermore, four types of GCNN models are proposed whose adjacency matrices are based on various bike-sharing system data, including Spatial Distance matrix (SD), Demand matrix (DE), Average Trip Duration matrix (ATD), and Demand Correlation matrix (DC). These six types of GCNN models and seven other benchmark models are built and compared on a Citi Bike dataset from New York City which includes 272 stations and over 28 million transactions from 2013 to 2016. Results show that the GCNNrec-DDGF performs the best in terms of the Root Mean Square Error, the Mean Absolute Error and the coefficient of determination (R2), followed by the GCNNreg-DDGF. They outperform the other models. Through a more detailed graph network analysis based on the learned DDGF, insights are obtained on the black box of the GCNN-DDGF model. It is found to capture some information similar to details embedded in the SD, DE and DC matrices. More importantly, it also uncovers hidden heterogeneous pairwise correlations between stations that are not revealed by any of those matrices.
View
... Probiotics such as Bacillus spp. Increased growth and are commonly used in livestock and poultry feed (Nayak 2010;Cruz et al., 2012;Gong et al., 2018;Wang et al., 2021) found that Bacillus species are helpful for lactic acid-producing bacteria (Leser et al., 2008). The growth of dangerous bacteria, such as salmonella, was considerably observed to be inhibited by intestinal environment acidity (Van Immerseel et al., 2006). ...
EFFECT OF LOW PROTEIN DIETS WITH SUPPLEMENTATION OF BACILLUS PUMILLUS ON GROWTH AND CARCASSES PERFORMANCE, HEMATOLOGICAL AND BIOCHEMICAL INDICES OF BROILER CHICKEN
Article
Full-text available
  • Jul 2024
The research objective was to determine the effect of Bacillus pumilus on growth performance hematological and biochemical indices of broilers chicken fed a low-protein diet. A total numbers of 180-day-old chicks were brought and placed into three groups, and each treatment was divided into six replicates (10 birds per replicate). Experimental diets included control (CO) having 22% crude protein (CP) for the starter and 20.0% CP for the finisher. However, the low protein (LP) contain 20% CP for the starter, 18.0% CP for the finisher, and Bacillus pumilus (BP-2000) treatment (LP+0.01% BP for starter and finisher). The birds fed a diet with BP-2000 improved the live body weight (LBW), and average daily gain (ADG) in the starter and overall period. In contrast, the feed conservation ratio (FCR) was reduced in the starter and overall phases (P<0.05). Dressing, breast, and leg muscle percentage significantly improved in BP-2000 (P< 0.05). Abdominal fat percentage was significantly reduced in BP-2000. The hemoglobin (HB), red blood cells (RBC), packed cell volume (PCV) showed no difference between all treatments, but the white blood cell (WBC) count was higher in BP-2000. Total protein (TP), albumin (ALB), globulin (GLB), and creatinine (CRAET) were not affected by dietary supplementation of BP-2000. High-density lipoprotein (HDL) was significantly higher in BP-2000 than control and LP groups (P<0.05). A lower serum level of total cholesterol (TC) and low-density lipoproteins (LDL) was observed in BP-2000 than LP. In conclusion, the current study showed that birds fed diet low in proteins with supplementation of BP-2000 improve growth performance, carcasses, and some blood biochemical indicators.
View
... While many studies have highlighted the effectiveness of LSTM in modeling the trajectory of individual vehicles, such as those by Lin et al. [24] and Kavran et al. [25], Vanilla LSTM models often overlook the interactions among surrounding vehicles. This limitation arises because RNN models excel primarily in learning temporal sequences. ...
A deep learning approach to predicting vehicle trajectories in complex road networks
Article
Full-text available
  • Jun 2024
Accurate prediction of vehicle trajectories is essential for safe and efficient navigation in urban environments, particularly with the increasing prevalence of autonomous vehicles and intelligent transportation systems. This paper introduces a deep learning-based approach for predicting vehicle trajectories on urban roads in real time. The method combines techniques from graph neural networks (GNNs) and long short-term memory (LSTM)-based models to capture intricate spatial and temporal dependencies among vehicles. Vehicles are represented as nodes in the proposed graph model, and graph attention mechanism is used to model the interactions between them. Additionally, LSTM modules encode motion patterns and temporal correlations, facilitating spatial and temporal information fusion to improve prediction accuracy. The effectiveness of the approach is demonstrated through extensive experimentation and evaluation in generating vehicle trajectories, surpassing baseline methods. The proposed method holds promise for real-time vehicle trajectory prediction, with the potential for applications in autonomous driving, traffic management, and intelligent transportation systems.
View
... The synthesis of control policies for heating, ventilation and air conditioning systems, autonomous vehicles, and robotics is a fundamental problem for cyber-physical systems (Lin et al., 2018;Wei et al., 2017;Xu et al., 2018). Researchers have demonstrated that reinforcement learning is effective at learning policies to achieve goal-reaching property to solve complex tasks, i.e. controlling the system to obtain the expected long-term reward (Dong et al., 2020;Haarnoja et al., 2018). ...
Safe reinforcement learning for dynamical systems using barrier certificates
Article
Full-text available
  • Dec 2022
Safety control is a fundamental problem in policy design. Basic reinforcement learning is effective at learning policy with goal-reaching property. However, it does not guarantee safety property of the learned policy. This paper integrates barrier certificates into actor-critic-based reinforcement learning methods in a feedback-driven framework to learn safe policies for dynamical systems. The safe reinforcement learning framework is composed of two interactive parts: Learner and Verifier. Learner trains the policy to satisfy goal-reaching and safety properties. Since the policy is trained on training datasets, the two properties may not be retained on the whole system. Verifier validates the learned policy on the whole system. If the validation fails, Verifier returns the counterexamples to Learner for retraining the policy in the next iteration. We implement a safe policy learning tool SRLBC and evaluate its performance on three control tasks. Experimental results show that SRLBC achieves safety with no more than 0.5× time overhead compared to the baseline reinforcement learning method, showing the feasibility and effectiveness of our framework.
View
... Such a technology may also provide real-time information about ongoing traffic conditions (Talebpour & Mahmassani, 2016), promote driving safety, cooperative driving and reducing total traffic fuel consumption (Vahidi & Sciarretta, 2018). To optimize the traffic flow of such vehicles (in the sense of road capacity, energy consumption, individual vehicle performance etc.) several models were suggested as control schemes (Anastassov et al., 2017;Gong et al., 2016;Lin et al., 2018;Mahmassani, 2016;Qin & Wang, 2018;Sun et al., 2018;Xu et al., 2020). Other models optimize the trajectories of a single AV in a linear portion of a road (He et al., 2015) or manage traffic trajectories where inter-vehicle conflicts pose a problem (Chai et al., 2017;Lu et al., 2019). ...
Control of autonomous vehicles flow using imposed speed profiles
Article
Full-text available
  • Sep 2021
The current non-connected autonomous vehicle scheme for speed changing along the road has limitations. Other alternatives require a central control or a complex communication system between the vehicles. We suggest a low cost and simple method for controlling the traffic of a line of autonomous vehicles using predetermined speed profiles imposed upon the vehicles along the road. We introduce a novel method to control autonomous vehicles traffic. Particularly, we investigate cases where specific velocities are required at some points along the road. This is done by comparing different velocity profiles for acceleration, deceleration or a combination of both. As traffic flow and speed limit may change due to upcoming road conditions, it is imperative to control vehicle line traffic, such that phantom jams will be prevented while preserving maximal traffic flow at minimum energy. We provide a comparison of these profiles for acceleration, deceleration, and for the combined case in terms of traffic flow, energy consumption, travel duration and the resulting jam characteristics. Following the comparison, we conclude that the best strategy would be to use linear speed profiles both to accelerate and to decelerate. Lastly, we suggest a tool to compare speed profiles for deceleration cases where the formation of an upstream propagated traffic congestion is inevitable.
View
... In the transportation domain, it was applied for vehicle trajectory prediction and driving behavior prediction. In [128], two types of deep learning models, i.e., Long Short-Term Memory (LSTM) and the Convolutional Neural Network (CNN) model, are proposed to predict a human-driven vehicles' trajectory. In particular, CNN feeds more information to the LSTM by using images, making trajectory prediction more accurate. ...
A Survey of Driving Safety With Sensing, Vehicular Communications, and Artificial Intelligence-Based Collision Avoidance
Article
  • Jun 2021
Accurately discovering hazards and issuing appropriate warnings to drivers in advance or performing autonomous control is the core of the Collision Avoidance (CA) system used to solve traffic safety problems. More comprehensive environmental awareness, diversified communication technologies, and autonomous control can make the CA system more accurate and effective, thereby improving driving safety. In addition, the assistance of Artificial Intelligence (AI) technology can make the CA system adapt to the environment and facilitate fast and accurate decisions. Considering the current lack of a thorough survey of driving safety with sensing, vehicular communications, and AI-based collision avoidance, in this paper, we survey existing researches for state-of-the-art data-driven CA techniques. Firstly, we discuss the major steps of CA and key research issues. For each step, we review the existing enabling techniques and research methods for CA in detail, including sensing and vehicular communication for safe driving, as well as CA algorithm design. Particularly, we present a comparison between the most common AI algorithms for different functions in the CA system. Testbeds and projects for CA are summarized next. Finally, several open challenges and future research directions are also outlined.
View
View
View
On the assessment of vehicle trajectory data accuracy and application to the Next Generation SIMulation (NGSIM) program data
Article
  • Dec 2011
  • TRANSPORT RES C-EMER
Trajectories drawn in a common reference system by all the vehicles on a road are the ultimate empirical data to investigate traffic dynamics. The vast amount of such data made freely available by the Next Generation SIMulation (NGSIM) program is therefore opening up new horizons in studying traffic flow theory. Yet the quality of trajectory data and its impact on the reliability of related studies was a vastly underestimated problem in the traffic literature even before the availability of NGSIM data. The absence of established methods to assess data accuracy and even of a common understanding of the problem makes it hard to speak of reproducibility of experiments and objective comparison of results, in particular in a research field where the complexity of human behaviour is an intrinsic challenge to the scientific method. Therefore this paper intends to design quantitative methods to inspect trajectory data. To this aim first the structure of the error on point measurements and its propagation on the space travelled are investigated. Analytical evidence of the bias propagated in the vehicle trajectory functions and a related consistency requirement are given. Literature on estimation/filtering techniques is then reviewed in light of this requirement and a number of error statistics suitable to inspect trajectory data are proposed. The designed methodology, involving jerk analysis, consistency analysis and spectral analysis, is then applied to the complete set of NGSIM databases.
View
Eevery New Car Connected by 2025 as Embedded Mobile Mobile Technology Drives Growth of Connected Car Market
  • Jan 2013
  • Gsma References
References GSMA, 2013. Eevery New Car Connected by 2025 as Embedded Mobile Mobile Technology Drives Growth of Connected Car Market. IEEE, 2012. https://www.ieee.org/about/news/2012/5september-2-2012.html (accessed 6.1.18).