Search

Path planning plays a significant role in robot navigation applications, as path exploration ability requires the knowledge of both the kinematics and the environments. Most of the current methods consider the planning process alone instead of combining the planning results with tracking control, which leads to a significant reduction in the availability of the path, especially in complex scenarios with missing GPS and low positioning sensor accuracy. This paper proposes a reinforcement learning-based path planning algorithm, which aims to consider the errors caused by the robot’s motion during the dead-reckoning process and effectively reduces the cumulative error within the optimization process. The simulation conclusion in the 2D scene verifies the effectiveness of the algorithm for reducing the cumulative error.

This paper presents a Simultaneous Localization And Mapping (SLAM) algorithm in the underwater environment. In this paper, forward-looking sonar is used to extract environmental features to establish a 2D grid map. The SLAM algorithm estimates the AUV’s (Autonomous Underwater Vehicle) pose by fusing multi-sensor data, including IMU data and the DVL data. To verify the algorithm, we simulated the experimental environment using the UUV-Simulator software. The results show that the algorithm can effectively suppress divergence and accurately locate the AUV.

Studies on wall-following problem mostly focus on an single robot only. This study proposes a wall-following method of multi-robot, which is based on moving target tracking. The leader robot moves along walls by using position information include angle and distance between robots and walls. The follower robot moves along the walls, tracks its leader and avoids collision at convex corner of walls. Two E-puck robots are used in experiment in this study. The results of experiment verify the feasibility of this method.

Recently, The devolopment of Autonomous Underwater Vehicle (AUV) have made significant progress, and AUV is widely used in marine resources exploration and underwater operations. In order to accurately perform the task, highprecision navigation and positioning technology are therefore essential. Current navigation methods rely on GPS navigation system, acoustic navigation system and inertial navigation system. However, GPS signals are affected by external environmental factors. Acoustic navigation system has the advantages of accuracy, but difficult to maintain the system recovery. Inertial navigation is not subject to interference from the environment but has accumulated errors. To address above issues, the topology factor is introduced which only considers the positional relationship between objects regardless of their shapes.

Recently, the developments of tracking systems have been significantly improved for ships tracking. Traditional approaches adopts a divide-and-conquer strategy, in which data association becomes quite challenging to collect right measurements in high density clutter environments. In this paper, a novel approach using Probability Hypothesis Density (PHD) filter is proposed for ships tracking, in which targets states are estimated based on set-valued measurements. Furthermore, the proposed solution also avoids the requirements of the prior parameters in the PHD filter, with respect to the birth intensities. During the tracking phase, the point matching method is also utilized to distinguish the ships and non-interested targets between consecutive frames, where the unchanged topology information is then utilized to initialize the birth intensities in the PHD filter.