Search Results

Recent researches in autonomous driving mainly consider the uncertainty in perception and prediction modules for safety enhancement. However, obstacles which block the field-of-view (FOV) of sensors could generate blind areas and leaves environmental uncertainty a remaining challenge for autonomous vehicles. Current solutions mainly rely on passive obstacles avoidance in path planning instead of active perception to deal with unexplored high-risky areas. In view of the problem, this paper introduces the concept of information entropy, which quantifies uncertain information in the blind area, into the motion planning module of autonomous vehicles. Based on model predictive control (MPC) scheme, the proposed algorithm can plan collision-free trajectories while actively explore unknown areas to minimize environmental uncertainty. Simulation results under various challenging scenarios demonstrate the improvement in safety and comfort with the proposed perception-aware planning scheme.

Multi-Target tracking is a central aspect of modeling the surrounding environment of autonomous vehicles. Automotive millimeter-wave radar is a necessary component in the autonomous driving system. One of the biggest advantages of radar is it measures the velocity directly. Another big advantage is that the radar is less influenced by environmental conditions. It can work day and night, in rainy or snowy conditions. In the expressway scenario, the forward-looking radar can generate multiple objects, to properly track the leading vehicle or neighbor-lane vehicle, a multi-target tracking algorithm is required. How to associate the track and the measurement or data association is an important question in a multi-target tracking system. This paper applies the nearest-neighbor method to solve the data association problem and uses an extended Kalman filter to update the state of the track.

Deep neural network models have been widely used for environment perception of intelligent vehicles. However, due to models’ innate probabilistic property, the lack of transparency, and sensitivity to data, perception results have inevitable uncertainties. To compensate for the weakness of probabilistic models, many pieces of research have been proposed to analyze and quantify such uncertainties. For safety-critical intelligent vehicles, the uncertainty analysis of data and models for environment perception is especially important. Uncertainty estimation can be a way to quantify the risk of environment perception. In this regard, it is essential to deliver a comprehensive survey. This work presents a comprehensive overview of uncertainty estimation in deep neural networks for environment perception of intelligent vehicles.

Aiming at the test safety problems in the early stage of self-driving cars development, firstly the virtual vehicle on-board CAN data acquisition module of the present project was designed based on virtual LabVIEW. Then a wireless remote control system for the self-driving car was constructed, which integrated the built virtual vehicle on-board CAN data acquisition system, the remote real-time image monitoring module and the remote upper computer control module based on ZigBee wireless transmission. It can execute the environmental awareness training and continuous and complex motion manipulation testing of the vehicle without relying on the driver, which can solve the safety problems in the tests of initial development of self-driving cars. Finally, the four-wheel independent steering electric vehicle was used as the self-driving test vehicle, and the wireless remote control system was tested on the double lane change type path and S-type path.

For high levels autonomous driving functions, the Decision Layer often takes on more responsibility due to the requirement of facing more diverse and even rare conditions. It is very difficult to accurately find a safe and efficient lane change timing when autonomous vehicles encounter complex traffic flow and need to change lanes. The traditional method based on rules and experiences has the limitation that it is difficult to be taken into account all possible conditions. Therefore, this paper designs a lane-changing decision algorithm based on data-driven and machine learning, and uses the DQN (Deep Q Network) algorithm in Reinforcement Learning to determine the appropriate lane-changing timing and target lane. Firstly, the scene characteristics of the highway are analyzed, the input and output of the decision-making model are designated and the data from the Perception Layer are processed.

The past decade has witnessed the rapid development of autonomous parking technology, since it has promising capacity to improve traffic efficiency and reduce the burden on drivers. However, it is prone to the trap of self-centeredness when each vehicle is automated parking in isolation. And it is easy to cause traffic congestion and even chaos when multiple autonomous vehicles require of parking into the same lot. In order to address the multiple vehicle parking problem, we propose a parking planning method with genetic algorithm. Firstly, an optimal mathematic model is established to describe the multiple autonomous vehicle parking problem. Secondly, a genetic algorithm is designed to solve the optimization problem. Thirdly, illustrative examples are developed to verify the parking planner. The performance of the present method indicates its competence in addressing parking multiple autonomous vehicles problem.

Although autonomous valet parking technology can replace the driver to complete the parking operation, it is easy to cause traffic chaos in the case of lacking scheduling for multiple parking agents, especially when multiple cars compete for the same parking slot at the same time. Therefore, in order to ensure orderly traffic and parking safety, it is necessary to allocate parking slots reasonably for multiple autonomous valet parking vehicles. The parking slots allocation model is built as an optimal problem with constraints. Both parking mileage cost and parking difficult cost are considering at the objective function in the optimization problem. There are three types of constraints. The first is the capacity limit of a single parking slot, the second is the space limit occupied by a single vehicle, and the third is the total capacity limit of the parking lot. After establishing parking slots allocation model, the immune algorithm is coded to solve the problem.

Lane marking detection plays a crucial role in Autonomous Driving Systems or Advanced Driving Assistance System. Vision based lane marking detection technology has been well discussed and put into practical application. LiDAR is more stable for challenging environment compared to cameras, and with the development of LiDAR technology, price and lifetime are no longer an issue. We propose a lane marking detection algorithm based on solid-state LiDARs. First a series of data pre-processing operations were done for the solid-state LiDARs with small field of view, and the needed ground points are extracted by the RANSAC method. Then, based on the OTSU method, we propose an approach for extracting lane marking points using intensity information.

With the development of autonomous driving, the penetration rate of autonomous vehicles on the road will continue to grow. As a result, the social cooperation ability of autonomous vehicles will have a great effect on the social acceptance of autonomous driving, which can be described as harmony with traffic. In order to research the evaluation method of the harmony with traffic, this paper proposes a subjective and objective mapping evaluation method based on the Mean Impact Value and Backpropagation (MIV-BP) Neural Network, with the merging vehicle on the expressway ramp as the research object. Firstly, by taking 16 original objective indexes obtained by theoretical analysis and the subjective evaluation results as input and output, respectively, the BP Neural Network model is constructed as a baseline model. Secondly, nine selected objective indexes are selected by the MIV method based on the baseline model.

Object detection is an important visual content of the autonomous vehicle, the traditional detecting methods usually cost a lot of computational memory and elapsed time. This paper proposes to use lightweight deep convolutional neural network (MobilenetV3-SSDLite) to carry out the object detection task of autonomous vehicles. Simulation analysis based on this method is implemented, the feature layer obtained after h-swish activation function in the first Conv of the 13th bottleneck module in MobilenetV3 is taken as the first effective feature layer, and the feature layer before pooling and convolution of the antepenultimate layer in MobilenetV3 is taken as the second effective feature layer, and these two feature layers are extracted from the MobilenetV3 network.

The main challenge for future autonomous vehicles is to identify their location and body pose in real time during driving, that is, “where am I? and how will I go?”. We address the problems of pose estimation and scene perception from continuous visual frames in intelligent vehicle. Recent advanced technology in the domain of deep learning proposes to train some learning models for vehicle’s series detection tasks in a supervised or unsupervised manner, which has numerous advances over traditional approaches, mainly reflected in the absence of manual calibration and synchronization of the camera and IMU. In the paper, we propose a novel approach for pose estimation and scene recognition with a deep fusion of multi-modal neural features in the manner of unsupervised. Firstly, low-cost camera and IMU are used to extract original visual and inertial data, then the visual and inertial encoders are utilized to encoder the feature of the two modes.

In the autonomous driving perception task, the accuracy of target detection is an essential evaluation, especially for small targets. In this work, we propose a multi-sensor fusion neural network that combines radar and image data to improve the confidence level of the camera when detecting targets and the accuracy of the prediction box regression. The fusion network is based on the basic structure of single-shot multi-box detection (SSD). Inspired by the attention mechanism in image processing, our work incorporates the a priori knowledge of radar detection in the convolutional block attention module (CBAM), which forms a new attention mechanism module called radar convolutional block attention module (RCBAM). We add the RCBAM into the SSD target detection network to build a deep neural network fusing millimeter-wave radar and camera.

Since situational awareness is crucial for autonomous driving in urban environments, multi-target tracking has become an increasingly popular research topic during the last several years. For autonomous driving in urban environments, cars and pedestrians are the two main types of obstacles, and their motion characteristics are not the same. While in the current related multi-target tracking research, the same motion model (such as Constant Velocity model [CV]) or motion model set (such as CV combined with Constant Acceleration model [CA]) is mostly used to track different types of obstacles simultaneously. Besides, in current research, regular motion models are mostly adopted to track pedestrians, such as CV, CA, and so on, the uncertainty in pedestrian motion is not well considered.

Since data processing methods could not completely eliminate the uncertainty of signals, it is a key issue for stable and robust decision-making for uncertainty tolerance of intelligent vehicles. In this paper, a decision-making for an Autonomous Emergency Braking (AEB) case considering the uncertainty of road adhesion coefficient estimation (RACE) is proposed. Firstly, the 3σ criterion is employed to classify the confidence in order to establish the decision-making mechanism considering the signal uncertainty of RACE. Secondly, the model for AEB with the uncertainty of the road adhesion coefficient estimated is designed based on the Seungwuk Moon model. Thirdly, a CCRs and CCRm scenario was designed to verify the feasibility in reference to the European New Car Assessment Programme (Euro NCAP) standard. Finally, the results of 10,000 cycles test illustrate that the proposed method is stable and could significantly improve the safety confidence both in the CCRs and CCRm scenarios.

Environmental perception is a crucial subsystem in autonomous vehicles. In order to build safe and efficient traffic transportation, several researches have been proposed to build accurate, robust and real-time perception systems. Camera and LiDAR are widely equipped on autonomous self-driving cars and developed with many algorithms in recent years. The fusion system of camera and LiDAR provides state-of the-art methods for environmental perception due to the defects of single vehicular sensor. Extrinsic parameter calibration is able to align the coordinate systems of sensors and has been drawing enormous attention. However, differ from spatial alignment of two sensors’ data, joint calibration of multi-sensors (more than two sensors) should balance the degree of alignment between each two sensors.

Tracking vehicle trajectories is essential for autonomous vehicles and advanced driver-assistance systems to understand traffic environment and evaluate collision risk. In order to reduce the position deviation and fluctuation of tracking on-road vehicle by millimeter-wave radar (MMWR), an interactive multi-model Kalman filter (IMM-KF) tracking algorithm including data association and track management is proposed. In general, it is difficult to model the target vehicle accurately due to lack of vehicle kinematics parameters, like wheel base, uncertainty of driving behavior and limitation of sensor’s field of view. To handle the uncertainty problem, an interacting multiple model (IMM) approach using Kalman filters is employed to estimate multitarget’s states. Then the compensation of radar ego motion is achieved, since the original measurement is under the radar polar coordinate system.

As a critical technology for autonomous vehicles, high precise positioning is essential for automated container terminals to implement intelligent dispatching and to improve container transport efficiency. Because of the unstable performance of global positioning system (GPS) in some circumstances, an ultra wide band (UWB) positioning system is developed for autonomous trucks in an automated container terminal. In this paper, an asynchronous structure is adopted in the system, and a three-dimensional (3D) localization method is proposed. Other than a traditional UWB positioning system with a server, in this asynchronous system, positions are calculated in the vehicle. Therefore, propagation delays from the server to vehicles are eliminated, and the real-time performance can be significantly improved. Traditional 3D localization methods based on time difference of arrival (TDOA) are mostly invalid with anchors in the same plane.

Safety is the cornerstone for Advanced Driver Assistance Systems (ADAS) and Autonomous Driving Systems (ADS). To assess the safety of a traffic situation, it is essential to predict motion states of traffic participants in the future with mathematic models. Accurate vehicle trajectory prediction is an important prerequisite for reasonable traffic situation risk assessment and appropriate decision making. Vehicle trajectory prediction methods can be generally divided into motion model based methods and maneuver model based methods. Vehicle trajectory prediction based on motion models can be accurate and reliable only in the short term. While vehicle trajectory prediction based on maneuver models present more satisfactory performance in the long term, these maneuver models rely on machine learning methods. Abundant data should be collected to train the maneuver recognition model, which increases complexity and lowers real-time performance.

Autonomous vehicle performance is increasingly highlighted in many highway driving scenarios, which leads to more priorities to vehicle stability as well as tracking accuracy. In this paper, a nonlinear model predictive controller for autonomous vehicle trajectory tracking is designed and verified through a real-time simulation bench of a virtual test track. The dynamic stability constraints of nonlinear model predictive control (NLMPC) are obtained by a novel quadrilateral stability region criterion instead of the conventional phase plane method using the double-line region. First, a typical lane change scene of overtaking is selected and a new composited trajectory model is proposed as a reference path that combines smoothness of sine wave and comfort of linear functional path. Reference lateral velocity, azimuth angle, yaw rate, and front wheel steering angle are subsequently taken into account.

A rapid path-planning algorithm that generates drivable paths for an autonomous vehicle operating in structural road is proposed in this paper. Cubic B-spline curve is adopted to generating smooth path for continuous curvature and, more, parametric basic points of the spline is adjusted to controlling the curvature extremum for kinematic constraints on vehicle. Other than previous approaches such as inverse kinematics, model-based prediction postprocess approach or closed-loop forward simulation, using the kinematics model in each iteration of path for smoothing and controlling curvature leading to time consumption increasing, our method characterized the vehicle curvature constraint by the minimum length of segment line, which synchronously realized constraint and smooth for generating path. And Differ from the path of robot escaping from a maze, the intelligent vehicle traveling on road in structured environments needs to meet the traffic rules.