Estimating a 2D homography from a pair of images is a fundamental task in computer vision. Contrary to most convolutional neural network-based homography estimation methods that use alternative four-point homography parameterization schemes, in this study, we directly estimate the $3\times 3$ homography matrix value. We show that after coordinate normalization, the magnitude difference and variance of the elements of the normalized $3\times 3$ homography matrix is very small. Accordingly, we present STN-Homography, a neural network based on spatial transformer network (STN), to directly estimate the normalized homography matrix of an image pair. To decrease the homography estimation error, we propose hierarchical STN-Homography and sequence STN-homography models in which the sequence STN-Homography can be trained in an end-to-end manner. The effectiveness of the proposed methods is demonstrated based on experiments on the Microsoft common objects in context (MSCOCO) dataset, and it is shown that they significantly outperform the current state-of-the-art. The average processing time of the three-stage hierarchical STN-Homography and the three-stage sequence STN-Homography models on a GPU are 17.85 ms and 13.85 ms, respectively. Both models satisfy the real-time processing requirements of most potential applications.

Vacuum suction units are widely used in various manufacturing lines, climbing robots, etc. Their most difficult problem is vacuum leakage, which leads to suction failure. Vacuum leakage is traditionally prevented by blocking the flow path between the atmosphere and the vacuum zone, which is difficult for a suction unit working on a rough surface. This paper proposes using the zero pressure difference (ZPD) method, which is based on a completely different mechanism. The ZPD method eliminates the pressure difference at the boundary of the vacuum zone, so vacuum leakage can be prevented regardless of the roughness of the working surface. A new vacuum suction unit based on the ZPD method was designed, fabricated, and tested. The ZPD suction unit forms a rotating water layer on the periphery of the vacuum zone, and the resulting inertial force generates a steep pressure gradient so that a high vacuum is maintained at the center of the vacuum zone while the pressure at the boundary remains equal to the atmospheric pressure. Experiments showed that a 0.8-kg ZPD suction unit generated a suction force of over 245 N on rough surfaces with a power consumption of less than 400 W. In contrast, a traditional suction unit of the same size would need a vacuum pump consuming several kilowatts and weighing dozens of kilograms to generate a similar suction force because of severe vacuum leakage. The ZPD suction unit was then successfully applied to a robotic arm, wall-climbing robot, and spider-man wall-climbing device.

This study presents a wall-climbing robot called Vortexbot. Vortexbot has a suction unit that uses vortex flow to generate a suction force. Unlike the traditional unit based on contact-type suction, the suction unit can maintain a suction force without any contact with the wall surface. Therefore, the suction unit can provide a climbing robot with sufficient stable suction force even on walls with very rough surfaces and raised obstacles/grooves, and there is no wear and tear. Furthermore, the compressed air vents from the gap between the suction unit and the wall surface after rotating in the vortex chamber. Hence, such kind of flow direction can avoid the effect of the dust and dropped items on the wall surface. In this paper, we first introduced the vortex suction unit principle and discuss the feasibility of its application to a wall-climbing robot. Subsequently, the mechanical structure of Vortexbot was designed. After which, we surveyed the suction properties of the suction unit on a smooth wall surface. Then the functional relationship between the percentage change in the suction force and the supply flow rate was obtained. In addition, we studied the effect of the roughness and shape (a raised obstacle and groove) of the wall surface on the suction performance of the suction unit. Finally, we experimentally verified the climbing performance of Vortexbot on several kinds of walls with different surface conditions. It was confirmed that using the suction unit improves the robot’s climbing performance.