Search Results

An autonomous vehicle is a comprehensive intelligent system that includes environment sensing, vehicle localization, path planning and decision-making control, of which environment sensing technology is a prerequisite for realizing autonomous driving. In the early days, vehicles sensed the surrounding environment through sensors such as cameras, radar, and lidar. With the development of 5G technology and the Vehicle-to-everything (V2X), other information from the roadside can also be received by vehicles. Such as traffic jam ahead, construction road occupation, school area, current traffic density, crowd density, etc. Such information can help the autonomous driving system understand the current driving environment more clearly. Vehicles are no longer limited to areas that can be sensed by sensors. Vehicles with different autonomous driving levels have different adaptability to the environment.

Effectively detecting road boundaries in real time is critical to the applications of autonomous vehicles, such as vehicle localization, path planning and environmental understanding. To precisely extract the road boundaries from the 3D-LiDAR data, a dedicated algorithm consisting of four steps is proposed in this paper. The steps are as follows: Firstly, the 3D-LiDAR data is pre-processed by employing the RANSAC method, the ground points are quickly separated from the original 3D-LiDAR point cloud to reduce the disturbance from the obstacles on the road, this greatly decreases the size of the point cloud to be processed. Secondly, based on the principle of 3D-LiDAR scanning, the ground points are divided into scan layers. And the road boundary points of each scan layer are detected by using three spatial features based on sliding window.

Lane detection and tracking play a key role in autonomous driving, not only in the LKA System but help estimate the pose of the vehicle. While there has been significant development in recent years, traditional outdoor SLAM algorithms still struggle to provide reliable information in challenging dynamic environments such as lack of roadside landscape or surrounding vehicles at almost the same speed or on the road in the woods. On the structured road, lane markings as static semantic features may provide a stable landmark assist in robust localization. As most of the current lane detection work mainly on separated images ignoring the relationship between adjacent frames, we propose a pixel-level lane tracking method for autonomous vehicles. In this paper, we introduce a deep network to detect and track lane features. The network has two parallel branches. One branch detects the lane position, while the other extracts the point description on a pixel level.

High-precision and real-time ego-motion estimation is vital for autonomous vehicle. There is a lot GPS-denied maneuver such as underground parking lot in urban areas. Therefore, the localization system relying solely on GPS cannot meets the requirements. Recently, lidar odometry and visual odometry have been introduced into localization systems to overcome the problem of missing GPS signals. Compared with visual odometry, lidar odometry is not susceptible to light, which is widely applied in weak-light environments. Besides, the autonomous parking is highly dependent on the geometric information around the vehicle, which makes building map of surroundings essential for autonomous vehicle. We propose a lidar inertial odometry and mapping. By sensor fusion, we compensate for the drawback of applying a single sensor, allowing the system to provide a more accurate estimate.

It’s hard to achieve complete self-driving using hand-crafting generalized decision-making rules, while the end-to-end self-driving system is low in complexity, does not require hand-crafting rules, and can deal with complex situations. Modular-based self-driving systems require multi-task fusion and high-precision maps, resulting in high system complexity and increased costs. In end-to-end self-driving, we usually only use camera to obtain scene status information, so image processing is very important. Numerous deep learning applications benefit from multi-task learning, as the multi-task learning can accelerate model training and improve accuracy with combine all tasks into one model, which reduces the amount of calculation and allows these systems to run in real-time. Therefore, the approach of obtaining rich scene state information based on multi-task learning is very attractive. In this paper, we propose an approach to multi-task learning for semantics, geometry and motion.

Adaptive Cruise Control (ACC) includes three modes: cruise control, car following control, and autonomous emergency braking. Among them, the car following control mode is mainly used to manage the speed and vehicle spacing approach the preceding vehicle within the range of smooth acceleration changes. In addition, although the motion information signal of the preceding vehicle can be collected by auxiliary equipment, it is still a random variable and normally regarded as a disturbance to affect the performance of vehicle controller. Therefore, this paper proposed an ACC strategy considering the disturbance of the preceding vehicle and multi-objective optimization.

Roundabouts are one of the most complex traffic environments in urban roads, and a key challenge for intelligent driving decision-making. Deep reinforcement learning, as an emerging solution for intelligent driving decisions, has the advantage of avoiding complex algorithm design and sustainable iteration. For the decision difficulty in roundabout scenarios, this paper proposes an artificial potential field based Soft Actor-Critic (APF-SAC) algorithm. Firstly, based on the Carla simulator and Gym framework, a reinforcement learning simulation system for roundabout driving is built. Secondly, to reduce reinforcement learning exploration difficulty, global path planning and path smoothing algorithms are designed to generate and optimize the path to guide the agent.

In the field of automatic driving, the combination of 3D LIDAR and inertial measurement unit (IMU) is a common sensor configuration scheme in laser point-cloud localization, high-precision map making and point-cloud target detection. So it is critical to calibrate LIDAR and IMU accurately. At present, due to the large volume and high cost of 3D LIDAR with high-line-number(Such as 64 lines or 128 lines), the configuration scheme of using multiple low-line-number 3D LIDARs appears in the automatic driving vehicle sensing system. However, the common calibration methods are not suitable for multi 3D LIDARs and IMU parameters calibration on autonomous vehicle, which have the disadvantages of cumbersome implementation and low accuracy. In this paper, a joint calibration test platform composed of dual LIDARs and IMU is assembled, and a method of precise automatic calibration based on GPS/RTK data is proposed.

Autonomous driving technology, as the product of the fifth stage of the information technology revolution, is of great significance for improving urban traffic and environmentally friendly sustainable development. Autonomous driving can be divided into three main modules. The input of the decision module is the perception information from the perception module and the output of the control strategy to the control module. The deep reinforcement learning method proposes an end-to-end decision-making system design scheme. This paper adopts the Deep Deterministic Policy Gradient Algorithm (DDPG) that incorporates the Priority Experience Playback (PER) method. The framework of the algorithm is based on the actor-critic network structure model. The model takes the continuously acquired perception information as input and the continuous control of the vehicle as output.

Autonomous driving in real-world urban traffic must cope with dynamic environments. This presents a challenging decision-making problem, e.g. deciding when to perform an overtaking maneuver or how to safely merge into traffic. The traditional autonomous driving algorithm framework decouples prediction and decision-making, which means that the decision-making and planning tasks will be carried out after the prediction task is over. The disadvantage of this approach is that it does not consider the possible impact of ego vehicle decisions on the future states of other agents. In this article, a decision-making and planning method which considers longitudinal interaction is represented. The method’s architecture is mainly composed of the following parts: trajectory sampling, forward simulation, trajectory scoring and trajectory selection. For trajectory sampling, a lattice planner is used to sample three-dimensionally in both the time horizon and the space horizon.

With the rapid development of modern economy and society, traffic congestion has become an increasingly serious problem. Vehicle cooperative driving can alleviate traffic congestion and improve road traffic capacity. Compare with vehicle separate control, cooperative driving combines various vehicle systems, and highly integrates information on obstacle location, vehicle status and driving intention. Then the controller uniformly issues instructions to ensure the orderly driving of the platoon. In the cooperative driving platoon, the displacement difference and the speed difference between vehicles have a certain relationship, which reduces the possibility of traffic accidents and then improves the safety of driving. In the process of cooperative driving, if there are multiple vehicles whose speeds don’t meet the current lane requirements, or if there are obstacles ahead, multi-vehicle lane change measures must be taken.

In the field of autonomous driving trajectory planning, it’s virtual to ensure real-time planning while guaranteeing feasibility and robustness. Current widely adopted approaches include decoupling path planning and velocity planning based on optimization method, which can’t always yield optimal solutions, especially in complex dynamic scenarios. Furthermore, search-based and sampling-based solutions encounter limitations due to their low resolution and high computational costs. This paper presents a novel spatio-temporal trajectory planning approach that integrates both search-based planning and optimization-based planning method. This approach retains the advantages of search-based method, allowing for the identification of a global optimal solution through search. To address the challenge posed by the non-convex nature of the original solution space, we introduce a spatio-temporal semantic corridor structure, which constructs a convex feasible set for the problem.