Search

In this paper, the problem of decision making for autonomous vehicles changing lanes is addressed by formulating multiple games in normal form for pairs of agents. This formulation generates the optimal action for the Ego vehicle at a given state and does not consider global optimality for all agents. The payoff matrices of the games are designed based on a user-defined set of rules. The constant parameters of these payoffs are then adjusted using neural learning to generate optimal behavior among the vehicles. An algorithm integrating deep reinforcement learning and game theory, regarded as Nash Q-learning, is included in the decision-making scheme. The applicability of the proposed method in a lane-changing scenario is tested via simulation. [ABSTRACT FROM AUTHOR]

Robust and adaptive vehicle state estimation and tracking algorithms have become a very important part within the autonomous driving field. The family of interacting multiple model (IMM) filters has shown to provide very effective and accurate state estimation in systems whose behavior patterns change significantly over time. The main reason for the improved performance of IMM filters compared to single model approaches is the mode mixing, which constantly aligns the different models. This paper proposes an innovative way for the mode mixing, when the state-vectors of the models are of different size. The proposed mixing approach consists of two parts: firstly mixing the common states and secondly weighting between original and mixed states based on the model probabilities. Results from artificial simulations and real world measurement setups are shown to demonstrate the validity of the approach. Compared to previously suggested solutions, the proposed approach is more general and the overall complexity of the mode mixing step is reduced, which is the main contribution of the presented paper. [ABSTRACT FROM AUTHOR]

Traffic awareness is the prerequisite of autonomous driving. Given the limitation of on-board sensors (e.g., precision and price), remote measurement from either infrastructure or other vehicles can improve traffic safety. However, the wireless communication carrying the measurement result undergoes fading, noise and interference and has a certain probability of outage. When the communication fails, the vehicle state can only be predicted by Bayesian filtering with a low precision. Higher communication resource utilization (e.g., transmission power) reduces the outage probability and hence results in an improved estimation precision. The power control subject to an estimate variance constraint is a difficult problem due to the complicated mapping from transmit power to vehicle-state estimate variance. In this paper, we develop an estimator consisting of several Kalman filters (KFs) or extended Kalman filters (EKFs) and an interacting multiple model (IMM) to estimate and predict the vehicle state. We propose to apply deep reinforcement learning (DRL) for the transmit power optimization. In particular, we consider an intersection and a lane-changing scenario and apply proximal policy optimization (PPO) and soft actor-critic (SAC) to train the DRL model. Testing results show satisfactory power control strategies confining estimate variances below given threshold. SAC achieves higher performance compared to PPO. [ABSTRACT FROM AUTHOR]

Autonomous vehicles will generate data from the variety of sensors they employ to track the surrounding environment. This data is inherently valuable, as it gives algorithm designers the potential to leverage prior experience in order to improve driving performance over time. This paper uses the lens of autonomous racing to provide an example of how data from previous iterations of driving can be used to improve quantitative metrics of performance. Two complementary algorithms are demonstrated in this paper. The first algorithm uses iterative learning control (ILC) to simultaneously improve lateral and longitudinal tracking of the desired racing trajectory over multiple laps, while the second algorithm is focused on altering the trajectory itself using a search method. When driven experimentally at the limits of handling, the result is a reduction in lap time of nearly 1.4 seconds, a major improvement. [ABSTRACT FROM AUTHOR]

GNSS/INS integrated solution has been extensively studied over the past decades. However, its performance relies heavily on environmental conditions and sensor costs. The GNSS positioning can obtain satisfactory performance in the open area. Unfortunately, its accuracy can be severely degraded in a highly urbanized area, due to the notorious multipath effects and none-line-of-sight (NLOS) receptions. As a result, excessive GNSS outliers occur, which causes a huge error in GNSS/INS integration. This paper proposes to apply a fish-eye camera to capture the sky view image to further classify the NLOS and line-of-sight (LOS) measurements. In addition, the raw INS and GNSS measurements are tightly integrated using a state-of-the-art probabilistic factor graph model. Instead of excluding the NLOS receptions, this paper makes use of both the NLOS and LOS measurements by treating them with different weightings. Experiments conducted in typical urban canyons of Hong Kong showed that the proposed method could effectively mitigate the effects of GNSS outliers, and an improved accuracy of GNSS/INS integration was obtained, when compared with the conventional GNSS/INS integration. [ABSTRACT FROM AUTHOR]

The localization system is one of the key components of autonomous vehicles. Widely-used GNSS (low-cost) alone can not meet the centimeter accuracy and is unstable in urban environment. Typically LiDAR-based localization system is very computationally demanding due to the usage of the dense point cloud map. In this paper, we propose a novel lightweight LiDAR-based localization system for autonomous vehicle in this paper. The proposed system only relies on lightweight poles and curbs landmark map, which is highly robust and efficient compared to other localization systems. Poles and curbs are selected as landmarks because of their commonality and stability. We novelly propose a Branch-and-Bound (BnB)-based global optimization method to tackle the data association problem of poles. Motion decoupling is adopted to decouple translation and rotation to improve the efficiency of the BnB-based algorithm. Besides, we propose a new local grid map-based representation for curbs to make better use of curb information. Cost functions for pole and curb are defined respectively and then fused for the subsequent non-linear optimization method to obtain the vehicle location. Experiments on KITTI dataset and our self-collected dataset demonstrate the efficiency and accuracy of our system. [ABSTRACT FROM AUTHOR]

The Internet of Vehicles (IoV) as a promising application of Internet of Things (IoT) has played a significant role in autonomous driving, by connecting intelligent vehicles. Autonomous driving needs to process the mass environmental sensing data in coordination with surrounding vehicles, and makes an accurate driving judgment accordingly. Since the vehicles always have limited computing resources, processing these data in parallel with efficient task scheduling is one of the most important topics. Most current work focuses on formulating special scenarios and service requirements as optimization problems. However, the complicated and dynamic environment of vehicular computing is hard to model, predict and control, making those previous methods unscalable and unable to reflect the real scenario. In this paper, a Multi-task Deep reinforcement learning approach for scalable parallel Task Scheduling (MDTS) is firstly devised. For avoiding the curse of dimensionality when coping with complex parallel computing environments and jobs with diverse properties, we extend the action selection in Deep Reinforcement Learning (DRL) to a multi-task decision, where the output branches of multi- task learning are fine-matched to parallel scheduling tasks. Child tasks of a job are accordingly assigned to distributed nodes without any human knowledge while the resource competition among parallel tasks is leveraged through shared neural network layers. Moreover, we design an appropriate reward function to optimize multiple metrics simultaneously, with emphasis on specific scenarios. Extensive experiments show that the MDTS significantly increases the overall reward compared with least- connection scheduling and particle swarm optimization algorithm from −16.71, −0.67 to 2.93, respectively. [ABSTRACT FROM AUTHOR]

Rear-end collision prediction has gained an increasing attention for safety improvement in smart cities. It is urgent to design efficient warning strategies for rear-end collisions which is one of the main causes of traffic accidents. The existing researches have been conducted to predict the collisions. Learning-based methods are proposed to solve this complicated issue, which the traditional methods are difficult to solve. However, due to some limitations in terms of the feature extraction and prediction performance, back-propagation learning methods are facing challenge. In this paper, we proposed a novel Rear-end Collision Prediction Mechanism with deep learning method (RCPM), in which a convolutional neural network model is established. In RCPM, the dataset is smoothed and expanded based on genetic theory to alleviate the class imbalance problem. The preprocessed dataset is divided into training and testing sets as the input to train our convolutional neural network model. The experimental results show that compared with the Honda, Berkeley and multi-layer perception neural-network-based algorithms, RCPM effectively improves performance to predict rear-end collisions. [ABSTRACT FROM AUTHOR]

Artificial intelligence (AI)-powered autonomous vehicles (AVs) can integrate different machine learning (ML) techniques to build up a complex autonomous driving system. However, single AV intelligence is not enough to cope with ever-changing driving environments. The underlying reason is that, with current neural network design and training algorithms, it is challenging for the driving model to generalize to diverse driving environments all at once due to sample inefficiency and the curse of dimensionality. Powerful computing resources and massive amount of data can be used to train a good driving model offline. However, the driving model obtained offline might fail in corner case scenarios. In this paper, we propose an intelligence networking framework among AVs assisted by multi-access edge computing (MEC) with end-to-end learning for demonstration. In this framework, driving road is divided into segments and data is collected for each road segment separately. Assisted by MEC networks, a continuously updated driving model is produced in near real-time for each road segment when the environment changes. By dividing the road into segments, we aim to reduce the burden of generalization since a single model only needs to adapt to a specific road segment. Simulation results show that our solutions can produce updated driving model for each road segment to adapt to environmental changes better than the existing scheme. Upcoming AVs can then adapt to changing environments by downloading the updated driving model. [ABSTRACT FROM AUTHOR]

The three main modules of autonomous vehicles, i.e., sensing, decision making, and motion controlling, have been studied separately in most existing works on autonomous driving, which overlook the correlations among these modules, leading to a result of unsatisfactory performance. In this paper, we propose a novel scheme that first tactfully processes the sensing data, then jointly learns and optimizes the decision-making and motion-controlling using reinforcement learning (RL). Specifically, the proposed scheme designs a novel state representation mechanism, where the sensing data goes through the attention layer and the convolutional neural network (CNN) layer sequentially. The attention layer focuses on extracting the most important local information and then CNN layer takes a broad view to comprehensively consider the global information for a better representation. Furthermore, the proposed scheme jointly learns decision-making and motion-controlling, therefore, the relevance of these two modules is implicitly considered, which helps to achieve a better autonomous driving policy. Extensive simulation results show that the proposed scheme is better than classic control methods and some RL methods in terms of safety, velocity, etc. We also demonstrate the respective functions of the attention layer and the CNN layer through ablation studies. Finally, we construct a traffic scene with a real autonomous vehicle, and verified the feasibility of the proposed scheme. [ABSTRACT FROM AUTHOR]

Unlike high-end three-dimensional (3-D) scanners with more than 16 layers which are mainly used in academia, low-end 3-D scanners with a few layers are being developed by sensor makers for installation in commercial advanced driver assistance system. The output of a low-end 3-D scanner is completely different from that of a full 3-D scanner and it is rather similar to the output of a 2-D scanner with a single layer. In this paper, a new framework for moving object detection and subsequent tracking using a low-end 3-D scanner with four layers is proposed. The proposed method uses the contours of the objects to obtain a robust association between a detection and a tracking. The proposed method comprises five steps: preprocessing, contour extraction, hypothesis generation, pruning, and moving object detection. In the preprocessing step, outliers, such as the ground or backlights from preceding vehicles, are removed and the scanned points are decomposed into segments, each of which corresponds to a single object. In the track hypothesis generation step, each segment is associated with an existing track maintained over multiple scans. The association method developed here uses the contour shape of the segments and is motivated by the linear programming and dynamic time warping. In the track hypothesis pruning step, unlikely tracks are removed from the hypothesis trees based on the proposed hypothesis scores. In the last step, moving objects are detected based on the track velocity. The proposed method is applied to four challenging real-world scenarios, and its validity is demonstrated via experimentation. [ABSTRACT FROM AUTHOR]

Predicting the future trajectories of surrounding agents has become an crucial problem to be solved for the safety of autonomous vehicles. Recent studies based on Long Short Term Memory (LSTM) networks have shown powerful abilities to model social interactions. However, many of these approaches focus on spatial interactions of the neighborhood agents but ignore temporal interactions that accompany spatial interactions. In this paper, we propose a Hierarchical Spatio-Temporal Attention architecture (HSTA), which activates the utilization of spatial interactions with different weights, and jointly considers the temporal interactions across time steps of all agents. More specially, the graph attention mechanism (GAT) is presented to capture spatial interactions, the multi-head attention mechanism (MHA) is conducted to encode temporal correlations of interactions and a state gated fusion (SGF) layer is used to integrate spatial and temporal interactions. We evaluate our proposed method against baselines on both pedestrian and vehicle datasets. The results show that our model is effective and achieves state-of-the-art achievements. [ABSTRACT FROM AUTHOR]

Predicting the state of dynamic objects in a real traffic environment is a key issue in autonomous driving vehicles. Various approaches have been proposed to learn the dynamics from visual observations with static background. However, minimal research has been conducted in a real traffic environment due to the complicated and changeable scenes. This paper proposes an adaptive multi-target future state prediction (position/velocity) method under autonomous driving conditions. In particular, an adaptive visual interaction method and control mechanism are introduced to overcome the change in the number of objects in continuous driving frames. In addition, a two-stream architecture with stage-wise learning is utilized for accurate object state prediction by simultaneously complementing spatial and temporal information. Experiments on two public challenging datasets, namely Udacity (CrowdAI) and Udacity (Autti), demonstrate the effectiveness of the proposed method on multi-target dynamic state prediction in a real traffic environment. [ABSTRACT FROM AUTHOR]

This paper presents a latent space reinforcement learning method for interpretable decision-making of autonomous vehicles at highway on-ramps. This method is based on the latent model and the combination model of the hidden Markov model and Gaussian mixture regression (HMM-GMR). It is difficult for the traditional decision-making method to understand the environment because its input is high-dimensional and lacks an understanding of the task. By utilizing the HMM-GMR model, we can obtain the interpretable state providing semantic information and environment understanding. A framework is proposed to unify representation learning with the deep reinforcement learning (DRL) approach, in which the latent model is used to reduce the dimension of interpretable state by extracting underlying task-relevant information. Experimental results are presented and the results show the right balance between driving safety and efficiency in the challenging scenarios of highway on-ramps merging. [ABSTRACT FROM AUTHOR]

Millimeter-wave radars are being increasingly integrated into commercial vehicles to support advanced driver-assistance system features. A key shortcoming for present-day vehicular radar imaging is poor azimuth resolution (for side-looking operation) due to the form factor limits on antenna size and placement. In this paper, we propose a solution via a new multiple-input and multiple-output synthetic aperture radar (MIMO-SAR) imaging technique, that applies coherent SAR principles to vehicular MIMO radar to improve the side-view (angular) resolution. The proposed 2-stage hierarchical MIMO-SAR processing workflow drastically reduces the computation load while preserving image resolution. To enable coherent processing over the synthetic aperture, we integrate a radar odometry algorithm that estimates the trajectory of ego-radar. The MIMO-SAR algorithm is validated by both simulations and real experiment data collected by a vehicle-mounted radar platform. [ABSTRACT FROM AUTHOR]

Driving a car in an urban setting is an extremely difficult problem, incorporating a large number of complex visual tasks; however, this problem is solved daily by most adults with little apparent effort. This paper proposes a novel vision-based approach to autonomous driving that can predict and even anticipate a driver's behavior in real time, using preattentive vision only. Experiments on three large datasets totaling over 200 000 frames show that our preattentive model can 1) detect a wide range of driving-critical context such as crossroads, city center, and road type; however, more surprisingly, it can 2) detect the driver's actions (over 80% of braking and turning actions) and 3) estimate the driver's steering angle accurately. Additionally, our model is consistent with human data: First, the best steering prediction is obtained for a perception to action delay consistent with psychological experiments. Importantly, this prediction can be made before the driver's action. Second, the regions of the visual field used by the computational model strongly correlate with the driver's gaze locations, significantly outperforming many saliency measures and comparable to state-of-the-art approaches. [ABSTRACT FROM PUBLISHER]

With the aims of safe, smart and sustainable future mobility, a personalized approach of trajectory planning and control based on user preferences is developed for lane-change of autonomous vehicles in this paper. First, a safe area during the lane change process is identified by using constraint Delaunay triangulation. Then, an improved rapidly-exploring Random Trees (i-RRT) is developed with B-spline to generate the feasible trajectory cluster, which is subject to the safe area boundaries and the vehicle dynamics. To extract a personalized trajectory from this cluster, we firstly adopt the fuzzy linguistic preference relation (FLPR) method to identify users’ preferences on driving, which can be reflected by their subjective objectives including driving safety, ride comfort and vehicle stability. Then, the technique for order preference by similarity to ideal situation (TOPSIS) is utilized to solve the multi-objective optimisation problem formulated by considering the user preferences. The algorithms proposed above are integrated, and both simulation and experimental validation are conducted under lane-change scenarios of autonomous driving. Simulation and experiment results show that proposed approach is able to successfully realize personalized trajectory planning and lane-change control, satisfying users’ various preferences and simultaneously ensure vehicle safety, demonstrating its feasibility and effectiveness. [ABSTRACT FROM AUTHOR]

Vehicle localization and mobility tracking are important tasks in autonomous driving. Traditional methods either have insufficient accuracy or rely on additional facilities to reach the desired accuracy for autonomous driving. In this paper, a multi-sensor multi-vehicle localization and mobility tracking framework is developed for autonomous vehicles equipped with GPS, inertial measurement unit (IMU), and an integrated sensing system. Our algorithm fuse the information from local onboard sensors as well as the observations of other vehicles or existing intelligent transportation system infrastructure such as road side units (RSU) to improve the precision and stability of localization and mobility tracking. Specifically, this framework incorporates the dynamic model of vehicles to achieve better localization and tracking performance. The communication delays during the information sharing process are explicitly taken into account in our algorithm development. Simulations manifest that not only the accuracy of localization and mobility tracking could be greatly enhanced in general, but also the robustness can be guaranteed under circumstances where traditional localization and tracking devices fail. [ABSTRACT FROM AUTHOR]

Active safety systems are of significant importance for autonomous vehicles operating in safety-critical situations like an obstacle-avoidance manoeuvre with high vehicle speed or poor road condition. However, a conventional electronic stability control system, may not always yield desired path following and yaw stability performance in such circumstances merely through brake intervention. This paper pursues a detailed investigation on utilising model predictive control (MPC) and torque vectoring for path following and yaw stability control of over-actuated autonomous electric vehicles (AEVs) in dangerous double lane change manoeuvre. The control problem of the AEV is formulated based on MPC by utilising active front steering and torque vectoring, and constraints are imposed explicitly on yaw rate and sideslip angle to ensure yaw stability. Four MPC-based controllers are designed based on double-track vehicle models. Specifically, they include two one-level controllers, i.e. one with torque vectoring and one with equal torque allocation, and two two-level controllers, i.e. one with optimisation-based torque allocation and one with rule-based allocation. These controllers are assessed extensively, with respect to passing velocity, tracking accuracy, tyre utilisation and robustness. The effect of horizon length on the control performance and computational efficiency is also investigated. [ABSTRACT FROM AUTHOR]

This paper introduces a method to extract driving behaviors from a human expert driver which are applied to an autonomous agent to reproduce proactive driving behaviors. Deep learning techniques were used to extract latent features from the collected data. Extracted features were clustered into behaviors and used to create velocity profiles allowing an autonomous driving agent could drive in a human-like manner. By using proactive driving behaviors, the agent could limit potential sources of discomfort such as jerk and uncomfortable velocities. Additionally, we proposed a method to compare trajectories where not only the geometric similarity is considered, but also velocity, acceleration and jerk. Experimental results in a simulator implemented in ROS show that the autonomous agent built with the driving behaviors was capable of driving similarly to expert human drivers. [ABSTRACT FROM AUTHOR]

With the rapid development of autonomous driving, collision avoidance has attracted attention from both academia and industry. Many collision avoidance strategies have emerged in recent years, but the dynamic and complex nature of driving environment poses a challenge to develop robust collision avoidance algorithms. Therefore, in this paper, we propose a decentralized framework named RACE: Reinforced Cooperative Autonomous Vehicle Collision AvoidancE. Leveraging a hierarchical architecture we develop an algorithm named Co-DDPG to efficiently train autonomous vehicles. Through a security abiding channel, the autonomous vehicles distribute their driving policies. We use the relative distances obtained by the opponent sensors to build the VANET instead of locations, which ensures the vehicle's location privacy. With a leader-follower architecture and parameter distribution, RACE accelerates the learning of optimal policies and efficiently utilizes the remaining resources. We implement the RACE framework in the widely used TORCS simulator and conduct various experiments to measure the performance of RACE. Evaluations show that RACE quickly learns optimal driving policies and effectively avoids collisions. Moreover, RACE also scales smoothly with varying number of participating vehicles. We further compared RACE with existing autonomous driving systems and show that RACE outperforms them by experiencing 65% less collisions in the training process and exhibits improved performance under varying vehicle density. [ABSTRACT FROM AUTHOR]

Deep reinforcement learning (DRL) has been successfully applied to end-to-end autonomous driving, especially in simulation environments. However, common DRL approaches used in complex autonomous driving scenarios sometimes are unstable or difficult to converge. This paper proposes two approaches to improve the stability of the policy model training with as few manual data as possible. For the first approach, reinforcement learning is combined with imitation learning to train a feature network with a small amount of manual data for parameters initialization. For the second approach, an auxiliary network is added to the reinforcement learning framework, which can leverage the real-time measurement information to deepen the understanding of environment, without any guide of demonstrators. To verify the effectiveness of these two approaches, simulations in image information-based and lidar information-based end-to-end autonomous driving systems are conducted, respectively. These approaches are not only tested in the virtual game world, but also applied in Gazebo, in which we build a 3D world based on the real vehicle model of Ranger XP900 platform, the real 3D obstacle model, and the real motion constraints with inertial characteristics, so as to ensure that the trained end-to-end autonomous driving model is more suitable for the real world. Experimental results show that the performance is increased by over 45% in the virtual game world, and can converge quickly and stably in Gazebo in which previous methods can hardly converge. [ABSTRACT FROM AUTHOR]

Nighttime driving is more dangerous than daytime driving-particularly for senior drivers. Three to four times as many driving-related deaths occur at night than in the daytime. To improve the safety of night driving, automatic pedestrian detection based on Infrared images has drawn increased attention because pedestrians tend to stand out more against the back- ground in infrared images than they do in visible light images. Nevertheless, pedestrian detection in infrared images is by no means trivial-many of the known difficulties carry over from visible light images, such as image variability occasioned by pedestrians being in different poses. Typically, several different pedestrian templates have to be used in order to deal with a range of poses. Furthermore, pedestrian detection is difficult because of poor infrared image quality (low resolution, low contrast, few distinguishable feature points, little texture information, etc.) and misleading signals. To address these problems, this paper introduces a shape-independent pedestrian-detection method. Our segmentation algorithm first estimates pedestrians' horizontal locations through projection-based horizontal segmentation and then determines pedestrians' vertical locations through brightness/bodyline-based vertical segmentation. Our classification method defines multidimensional histogram-, inertia-, and contrast-based classification features. The features are shape-independent, complementary to one another, and capture the statistical similarities of image patches containing pedestrians with different poses. Thus, our pedestrian-detection system needs only one pedestrian template-corresponding to a generic walking pose-and avoids brute-force searching for pedestrians throughout whole im- ages, which typically involves brightness-similarity comparisons between candidate image patches and a multiplicity of pedestrian templates. Our pedestrian-detection system is neither based on tracking nor does it depend on. camera calibration to determine the relationship between an object's height and its vertical image locations. Thus, it is less restricted in applicability. Even if much work is still needed to bridge the gap between present pedestrian-detection performance and the high reliability required for real-world applications, our pedestrian-detection system is straightforward and provides encouraging results in improving speed, reliability, and simplicity. [ABSTRACT FROM AUTHOR]

Current research on Deep ReinforcementLearning (DRL) for automated on-ramp merging neglects vehicle powertrain and dynamics. This work considers automated on-ramp merging for a power-split Plug-In Hybrid Electric Vehicle (PHEV), the 2015 Toyota Prius Plug-In, using DRL. The on-ramp merging control and the PHEV energy management are co-optimized such that the DRL policy directly outputs the power split between the engine and the electric motor. The testing results show that DRL can be successfully used for co-optimization, leading to collision-free on-ramp merging. When compared with sequential approaches wherein the upper-level on-ramp merging control and the lower-level PHEV energy management are performed independently and in sequence, we found that co-optimization results in economic but jerky on-ramp merging while sequential approaches may result in collisions due to neglecting powertrain power limit constraints in designing the upper-level on-ramp merging controller. [ABSTRACT FROM AUTHOR]

In this study, a decision-making and motion planning controller with continuous action space is constructed in the highway driving scenario based on deep reinforcement learning. In the decision-making and planning problem, the goal is to achieve the safety, efficiency, and comfort of automated vehicles. In the driving scenario, the surrounding vehicles are controlled by the intelligent driver model and a general model (minimizing overall braking induced by lane change, MOBIL), which enables them to react to the environment and mimic the vehicle interactions on the highway. Given the uncertainties in the driving conditions, a specific deep reinforcement learning technique, called soft actor-critic, is used to solve the decision-making and planning problem with continuous action space. Simulation results show that the proposed method can solve the decision-making and motion planning problem in the interactive traffic environment to carry out safe lane-change maneuvers and cruise at high speed. In addition, two control policies are developed with different weights on safety, efficiency, and comfort. [ABSTRACT FROM AUTHOR]

A novel framework is proposed for enhancing the driving safety and fuel economy of autonomous vehicles (AVs) with the aid of vehicle-to-infrastructure (V2I) communication networks. The problem of driving trajectory design is formulated for minimizing the total fuel consumption, while enhancing driving safety (by obeying the traffic rules and avoiding obstacles). In an effort to solve this pertinent problem, a deep reinforcement learning (DRL) approach is proposed for making collision-free decisions. Firstly, a deep Q-network (DQN) aided algorithm is proposed for determining the trajectory and velocity of the AV by receiving real-time traffic information from the base stations (BSs). More particularly, the AV acts as an agent to carry out optimal action such as lane change and velocity change by interacting with the environment. Secondly, to overcome the large overestimation of action values by the Q-learning model, a double deep Q-network (DDQN) algorithm is proposed by decomposing the max-Q-value operation into action selection and action evaluation. Additionally, three practical driving policies are also proposed as benchmarks. Numerical results are provided for demonstrating that the proposed trajectory design algorithms are capable of enhancing the driving safety and fuel economy of AVs. We demonstrate that the proposed DDQN based algorithm outperforms the DQN based algorithm. Additionally, it is also demonstrated that the proposed fuel-economy (FE) based driving policy derived from the DRL algorithm is capable of achieving in excess of 24% of fuel savings over the benchmarks. [ABSTRACT FROM AUTHOR]

Autonomous driving has attracted significant attention in recent years. With the booming of artificial intelligence (AI), deep learning technologies have been applied to autonomous driving to help vehicles better perceive the environment. Besides the perceiving environment, predictive driving is another prominent smooth control and safe driving skill for human drivers. In this work, we develop a deep Monte Carlo Tree Search (deep-MCTS) control method for vision-based autonomous driving. Compared with existing deep learning-based autonomous driving control methods, our method can predict driving maneuvers to help improve the stability and performance of driving control. Two deep neural networks (DNNs) are employed for predicting action-state transformation and obtaining action-selection probabilities, respectively. The deep-MCTS utilizes the predicted information of the two DNNs and reconstructs multiple possible trajectories to predict driving maneuvers. An optimal trajectory is selected by the deep-MCTS based on both current road conditions and predicted driving maneuvers. The proposed method achieves high control stability by avoiding sharp turns and driving deviations. We implement our algorithm in the Udacity and Torcs self-driving environments. The testing results show that our algorithm achieves a significant improvement in training efficiency, the stability of steering control, and stability of driving trajectory compared to existing methods. [ABSTRACT FROM AUTHOR]

Lane detection in driving scenes is an important module for autonomous vehicles and advanced driver assistance systems. In recent years, many sophisticated lane detection methods have been proposed. However, most methods focus on detecting the lane from one single image, and often lead to unsatisfactory performance in handling some extremely-bad situations such as heavy shadow, severe mark degradation, serious vehicle occlusion, and so on. In fact, lanes are continuous line structures on the road. Consequently, the lane that cannot be accurately detected in one current frame may potentially be inferred out by incorporating information of previous frames. To this end, we investigate lane detection by using multiple frames of a continuous driving scene, and propose a hybrid deep architecture by combining the convolutional neural network (CNN) and the recurrent neural network (RNN). Specifically, information of each frame is abstracted by a CNN block, and the CNN features of multiple continuous frames, holding the property of time-series, are then fed into the RNN block for feature learning and lane prediction. Extensive experiments on two large-scale datasets demonstrate that, the proposed method outperforms the competing methods in lane detection, especially in handling difficult situations. [ABSTRACT FROM AUTHOR]