Visual SLAM has regained attention due to its ability to provide perceptual capabilities and simulation test data for Embodied AI. However, traditional SLAM methods struggle to meet the demands of high-quality scene reconstruction, and Gaussian SLAM systems, despite their rapid rendering and high-quality mapping capabilities, lack effective pose optimization methods and face challenges in geometric reconstruction. To address these issues, we introduce FGO-SLAM, a Gaussian SLAM system that employs an opacity radiance field as the scene representation to enhance geometric mapping performance. After initial pose estimation, we apply global adjustment to optimize camera poses and sparse point cloud, ensuring robust tracking of our approach. Additionally, we maintain a globally consistent opacity radiance field based on 3D Gaussians and introduce depth distortion and normal consistency terms to refine the scene representation. Further-more, after constructing tetrahedral grids, we identify level sets to directly extract surfaces from 3D Gaussians. Results across various real-world and large-scale synthetic datasets demonstrate that our method achieves state-of-the-art tracking accuracy and mapping performance.

Ground segmentation is an important preprocessing task for autonomous vehicles (AVs) with 3D LiDARs. However, the existing ground segmentation methods are very difficult to balance accuracy and computational complexity. This paper proposes a fast point cloud ground segmentation approach based on a coarse-to-fine Markov random field (MRF) method. The method uses the coarse segmentation result of an improved local feature extraction algorithm instead of prior knowledge to initialize an MRF model. It provides an initial value for the fine segmentation and dramatically reduces the computational complexity. The graph cut method is then used to minimize the proposed model to achieve fine segmentation. Experiments on two public datasets and field tests show that our approach is more accurate than both methods based on features and MRF and faster than graph-based methods. It can process Velodyne HDL-64E data frames in real-time (24.86 ms, on average) with only one thread of the I7-8700 CPU. Compared with methods based on deep learning, it has better environmental adaptability.

Effective semantic segmentation of lane marking is crucial for construction of high-precision lane level maps. In recent years, a number of different methods for semantic segmentation of images have been proposed. These methods concentrate mainly on analysis of camera images, due to limitations with the sensor itself, and thus far, the accurate three-dimensional spatial position of the lane marking could not be obtained, which hinders lane level map construction.This article proposes a lane marking semantic segmentation method based on LIDAR and camera image fusion using a deep neural network. In the approach, the object of the semantic segmentation is a bird’s-eye view converted from a LIDAR points cloud instead of an image captured by a camera. First, the DeepLabV3+ network image segmentation method is used to segment the image captured by the camera, and the segmentation result is then merged with the point clouds collected by the LIDAR as the input of the proposed network. A long short-term memory (LSTM) structure is added to the neural network to assist the network in semantic segmentation of lane markings by enabling use of time series information. Experiments on datasets containing more than 14,000 images, which were manually labeled and expanded, showed that the proposed method provides accurate semantic segmentation of the bird’s-eye view LIDAR points cloud. Consequently, automation of high-precision map construction can be significantly improved. Our code is available athttps://github.com/rolandying/FusionLane.

An effective and accurate lane marking detection algorithm is a fundamental element of the intelligent vehicle system and the advanced driver assistant system, which can provide important information to ensure the vehicle runs in the lane or warn the driver in case of lane departure. However, in the complex urban environment, lane markings are always affected by illumination, shadow, rut, water, other vehicles, abandoned old lane markings and non-lane markings, etc. Meanwhile, the lane markings are weak caused by hard use over time. The dash and curve lane marking detection is also a challenge. In this paper, a new lane marking detection algorithm for urban traffic is proposed. In the low-level phase, an iterative adaptive threshold method is used for image segmentation, which is especially suitable for the blurred and weakened lane markings caused by low illumination or wear. In the middle-level phase, the algorithm clusters the candidate pixels into line segments, and the upper and lower structure is used to cluster the line segments into candidate lanes, which is more suitable for curve and dashed lane markings. In the high-level phase, we compute the highest scores to get the two optimal lane markings. The optimal strategy can exclude interference similar to lane markings. We test our algorithm on Future Challenge TSD-Lane dataset and KITTI UM dataset. The results show our algorithm can effectively detect lane markings under multiple disturbance, occlusions and sharp curves.

This paper proposes an online reference path correction algorithm for Autonomous Ground Vehicles (AGVs) driving in unstructured environments. Because the reference path which is not in accord with the road geometry cannot be directly executed by AGV, we use a correction baseline, which represents the road geometry, as the guidance to correct reference path. As the road geometry varies according to the road environment, we use the centerline of drivable area extracted with General Regression Neural Network (GRNN) or optimal path generated by multi-resolution local trajectory planner as the correction baseline to correct the distance and curvature errors of the reference path. The proposed algorithm is tested on our experimental autonomous vehicle in the realistic off-road scenarios. Experimental results demonstrated its capability and efficiency to reduce the search space of the local trajectory planner and improve the robustness of tracking the reference path.

The acquisition, presentation and management of autonomous driving decision-making knowledge of unmanned vehicles are the key and difficult issues in the autonomous driving decision-making system of unmanned vehicles. This paper presents a knowledge model, which includes problem description layer and problem-solving knowledge layer. The automatic driving decision knowledge base of unmanned vehicle is composed of a set of knowledge models. Knowledge model supports knowledge representation and reasoning. Based on the WEB visualization knowledge modeling tool and visualization knowledge service tool, we construct the decision-making knowledge base management system for autonomous driving of unmanned vehicles and then construct the autonomous driving decision-making system of unmanned vehicles. The reasoning example shows that the knowledge base management system can effectively improve the knowledge acquisition, representation and maintenance efficiency of autonomous driving decision-making system, which is of great significance in enhancing the intelligence level of autonomous driving decision-making system.

As autonomous-vehicle-related technologies tend to be mature, improving passengers' experience by learning driving styles from human drivers becomes a promising research topic. This study aims at learning human drivers' velocity planning strategies for driving at curvy paths (e.g. negotiating sharp curves, turning at intersections, etc.) on structural road. First, we identified and extracted training trips from the latest naturalistic driving study database. Vehicle trajectories and the disturbances caused by other vehicles were estimated based on sensor data. Road characteristics, environmental parameters were identified from road information database and video clips. Then, neural network based models were developed to fit drivers' speed profiles under different driving situations. Five models with different prediction steps were trained by up to 600 driving trips. Three error criteria were used to evaluate the performance of proposed models. This study verified the possibility of using human drivers' experience to generate velocity recommendations for different driving conditions. The limitations of the models are also documented.

Publishing and delivering traffic information generally depend on message board and public radio service in China, which usually sends out static information without dynamic interaction between the services and the users. With the development of WebGIS and its successful application, it is possible to introduce the Web-based multiple architecture into the traffic information system. Currently, most WebGIS portals only provide static information in words and figures without any significant improvement in dynamic aspect, which is a very important requirement for traffic information system. We develop an urban traffic information portal for dynamic information updating and responding, which is based on the fast growing rich Internet application and WebGIS technology. The portal can not only provide the real-time information but also can interact with users for more specific. This paper introduces the multi-layer architecture used in our project. Reusable services related with GIS services are designed and encapsulated as portlet. We briefly introduce some implements and optimization at the end of this paper.

Most of the current semantic representation methods for service capabilities are usually based on top-down methodology. We aim to develop services characterization methods with statistical study on existing Web services and to improve services capability representation with bottom-up software services comprehension. Two services characterization methods are proposed in our work: quantitative statistical study which used for probing the distribution of main objects in Web services and relational statistical study which used for clustering actions or contents and measuring the similarity of Web services. A statistical relational model is proposed for mining and recognizing the patterns of services.

Provides an experimental study of planar impact of a robot manipulator on a surface. Using the data collected from a series of experiments, it investigates post-impact behavior for different pre-impact conditions such as configuration of the robot, angle and velocity of impact, etc. Understanding the post-impact behavior for various pre-impact conditions can result in the design of improved transition control strategies to stabilize the manipulator onto the surface. Potential applications of this study include robotic surface finishing operations such as polishing, chamfering, deburring, and grinding, etc.

This paper addresses the problem of contact transition from free motion to constrained motion for geometrically constrained robots. Constraint uncertainty can cause the robot to impact the surface with a non-zero velocity. To deal with contact transition. A new stable discontinuous transition controller is proposed. The control algorithm for a complete robot task that involves both free motion and constrained motion is also developed. Extensive experiments with the proposed control strategy were conducted with different levels of constraint uncertainty. Experimental results show much improved transition performance and force regulation. Details of the experimental platform and typical experimental results are given.

In this work, we consider adaptive control of time-varying mechanical systems. First, we derive the dynamics of time-varying mechanical systems under the assumption that the generalized constraints on the system do not depend on time but the system parameters such as masses and payloads are time-varying. A compact parameterization of the dynamics of the system is obtained based on the rate of change of parameters being a polynomial in time. Based on this parameterization, adaptation laws for the time-varying parameters is proposed. To test the proposed adaptive controllers, an experimental platform consisting of a two-link robot with a time-varying payload is designed. The designed experimental platform mimics pouring/filling operations in industry, where the payload is time-varying. Experimental results demonstrate the effectiveness of the proposed adaptive control designs.