Local Homing Navigation Based on the Moment Model for Landmark Distribution and Features

Abstract

For local homing navigation, an agent is supposed to return home based on the surrounding environmental information. According to the snapshot model, the home snapshot and the current view are compared to determine the homing direction. In this paper, we propose a novel homing navigation method using the moment model. The suggested moment model also follows the snapshot theory to compare the home snapshot and the current view, but the moment model defines a moment of landmark inertia as the sum of the product of the feature of the landmark particle with the square of its distance. The method thus uses range values of landmarks in the surrounding view and the visual features. The center of the moment can be estimated as the reference point, which is the unique convergence point in the moment potential from any view. The homing vector can easily be extracted from the centers of the moment measured at the current position and the home location. The method effectively guides homing direction in real environments, as well as in the simulation environment. In this paper, we take a holistic approach to use all pixels in the panoramic image as landmarks and use the RGB color intensity for the visual features in the moment model in which a set of three moment functions is encoded to determine the homing vector. We also tested visual homing or the moment model with only visual features, but the suggested moment model with both the visual feature and the landmark distance shows superior performance. We demonstrate homing performance with various methods classified by the status of the feature, the distance and the coordinate alignment.

Bulletin of the Polish Academy of Sciences: Technical Sciences

“…7. (4,1), (9,5) and (13,4) under the original version were separately selected as destinations, which are discretely distributed in the experiment scene. Figure 8 plots a representation of the AE grids corresponding to the home vector fields, overall darker plots represent better effects.…”

“…Figure 9 shows the RR results of MLBH and SL-ALV. The home locations are still set to (4,1) (9,5) and (13,4). For each home location, we select all of the six image versions to perform the experiments.…”

“…Visual homing is an attractive technology in this field, which provides nodeto-node navigation solutions for a mobile robot to return from the current location to the reference home location. The robot compares the currently viewed image with the pre-stored goal image to generate the home vector, which describes the moving course of the robot and guides the robot to move towards the home location [4][5][6].…”

Autonomous Shepherding Behaviors of Multiple Target Steering Robots

“…The localization information of agents is obtained using range sensors [ 25 ], overhead camera [ 26 ], laser sensors [ 37 , 38 ] or wireless motion-tracking sensors [ 32 , 36 ]. Many navigation systems have a target, which a mobile agent is supposed to reach with various sensor readings [ 37 , 38 ], whereas in the shepherding task, only steering agents have target information and make an effort to guide sheep agents. Then, the observers or steering agents produce appropriate actions to control the sheep robots to form a flock or move the group of sheep robots towards the target position.…”

Bulletin of the Polish Academy of Sciences: Technical Sciences

“…HiSS adopts the SIFT features as the landmarks, and classifies all the features into contracted features and expanded features by the relation between the SIFT scale value and spatial distance. These two different types of the features can produce the landmark vectors in different directions, and the desired home vector can be generated by integrating all the landmark vectors [24,25].…”