Autonomous vehicles can achieve accurate localization and real-time road information perception using sensors such as global navigation satellite systems (GNSSs), light detection and ranging (LiDAR), and inertial measurement units (IMUs). With road information, vehicles can navigate autonomously to a given position without traffic accidents. However, most of the research on autonomous vehicles has paid little attention to road profile information, which is a significant reference for vehicles driving on uneven terrain. Most vehicles experience violent vibrations when driving on uneven terrain, which reduce the accuracy and stability of data obtained by LiDAR and IMUs. Vehicles with an active suspension system, on the other hand, can maintain stability on uneven roads, which further guarantees sensor accuracy. In this paper, we propose a novel method for road profile estimation using LiDAR and vehicles with an active suspension system. In the former, 3D laser scanners, IMU, and GPS were used to obtain accurate pose information and real-time cloud data points, which were added to an elevation map. In the latter, the elevation map was further processed by a Kalman filter algorithm to fuse multiple cloud data points at the same cell of the map. The model predictive control (MPC) method is proposed to control the active suspension system to maintain vehicle stability, thus further reducing drifts of LiDAR and IMU data. The proposed method was carried out in outdoor environments, and the experiment results demonstrated its accuracy and effectiveness.

As a complex task, robot sorting has become a research hotspot. In order to enable robots to perform simple, efficient, stable and accurate sorting operations for stacked multi-objects in unstructured scenes, a robot multi-object sorting system is built in this paper. Firstly, the training model of rotating target detection is constructed, and the placement state of five common objects in unstructured scenes is collected as the training set for training. The trained model is used to obtain the position, rotation angle and category of the target object. Then, the instance segmentation model is constructed, and the same data set is made, and the instance segmentation network model is trained. Then, the optimized Mask R-CNN instance segmentation network is used to segment the object surface pixels, and the upper surface point cloud is extracted to calculate the normal vector. Then, the angle obtained by the normal vector of the upper surface and the rotation target detection network is fused with the normal vector to obtain the attitude of the object. At the same time, the grasping order is calculated according to the average depth of the surface. Finally, after the obtained object posture, category and grasping sequence are fused, the performance of the rotating target detection network, the instance segmentation network and the robot sorting system are tested on the established experimental platform. Based on this system, this paper carried out an experiment on the success rate of object capture in a single network and an integrated network. The experimental results show that the multi-object sorting system based on deep learning proposed in this paper can sort stacked objects efficiently, accurately and stably in unstructured scenes.

There are few studies on the intelligent guidance of unmanned sailboats, which should coordinate pluralistic tasks at sea in the nature of its maneuvring intractability. To ensure the algorithmic practicability, this paper proposes a path-following and collision-avoidance guidance approach of unmanned sailboats with total formulaic description. The risk-detecting mechanism is fabricated by setting a circular detecting zone and using the time to the closest point of approach. Then, the risk of collision, the path deviation, the speed loss, and the course loss can be judged by constructing the cost functions and applying the distance to closest point of approach. The optimized heading angle is deemed as the one minimizing the aggregate cost functions, which is sought by applying and improving the beetle antennae search (BAS) algorithm. In the proposed modified BAS, the searching step is redesigned to enhance the searching efficiency. To ensure the convergence of the real heading angle to the reference, the backstepping-based control law is fabricated for the high-order sailboat model and in the linear form. The control parameters are offline optimized through the modified BAS. Compared with the adaptive control, this controller can guarantee more computation simplicity and the optimized control performance. Finally, simulation corroborates that the sailboat can successfully complete path following and collision avoidance while encountering multiple static and moving obstacles under the proposed schemes.

Autonomous vehicles can obtain real-time road information using 3D sensors. With road information, vehicles avoid obstacles through real-time path planning to improve their safety and stability. However, most of the research on driverless vehicles have been carried out on urban even driveways, with little consideration of uneven terrain. For an autonomous full tracked vehicle (FTV), the uneven terrain has a great impact on the stability and safety. In this paper, we proposed a method to predict the pose of the FTV based on accurate road elevation information obtained by 3D sensors. If we could predict the pose of the FTV traveling on uneven terrain, we would not only control the active suspension system but also change the driving trajectory to improve the safety and stability. In the first, 3D laser scanners were used to get real-time cloud data points of the terrain for extracting the elevation information of the terrain. Inertial measurement units (IMUs) and GPS are essential to get accurate attitude angle and position information. Then, the dynamics model of the FTV was established to calculate the vehicle’s pose. Finally, the Kalman filter was used to improve the accuracy of the predicted pose. Compared to the traditional method of driverless vehicles, the proposed approach was more suitable for autonomous FTV. The real-world experimental result demonstrated the accuracy and effectiveness of our approach.

This research develops a gravity compensation method that determines the mass of a task object easily and compensates for the external force caused by the task object when it is conveyed by a hydraulic teleoperation construction robot. Moreover, this study establishes a master–slave system for this robot; two joysticks act as the master, and an excavator with four links (fork glove, swing, boom, and arm) represents the slave. To compensate for the influence of gravity, a previous gravity compensation method is proposed and applied to the boom and arm. However, it is ineffective during the conveyance process especially when the task object is heavy because the driving force is influenced by gravity of the task object. Therefore, this research presents a gravity compensation method that can effectively determine the mass of a grasped object and compensate for the external force induced by its gravity, as verified through pressing, grasping, and conveying experiments.