This letter introduces a novel monocular visual odometry network structure, leveraging the Swin Transformer as the backbone network, named SWformer-VO. It can directly estimate the six degrees of freedom camera pose under monocular camera conditions by utilizing a modest volume of image sequence data with an end-to-end methodology. SWformer-VO introduces an Embed module called “Mixture Embed”, which fuses consecutive pairs of images into a single frame and converts them into tokens passed into the backbone network. This approach replaces traditional temporal sequence schemes by addressing the problem at the image level. Building upon this foundation, various parameters of the backbone network are continually improved and optimized. Additionally, experiments are conducted to explore the impact of different layers and depths of the backbone network on accuracy. Excitingly, on the KITTI dataset, SWformer-VO demonstrates superior accuracy compared with common deep learning-based methods such as SFMlearner, Deep-VO, TSformer-VO, Depth-VO-Feat, GeoNet, Masked Gans and others introduced in recent years. Moreover, the effectiveness of SWformer-VO is also validated on our self-collected dataset consisting of nine indoor corridor routes for visual odometry.

Nano-positioning is widely used in Micro-electromechanical Systems (MEMS), micromanipulator and biomedicine, coupling errors and tiny output displacements are the main disadvantages of the one. A totally uncoupled micro/nano-positioning stage with lever amplifiers is designed and tested in this paper. It is fully symmetrical along with the x- and y-directions. For obtaining large output displacements, two fully symmetric two-stage lever displacement amplifiers are utilized to amplify output displacements of piezoelectric actuators (PZTs). The established models for performances evaluation of the stage, in terms of kinetostatics, amplification ratio, reachable workspace, the input and output stiffness, are verified by finite element analysis (FEA). After that, the dimensional optimization is also carried out through the genetic optimization algorithm.The prototype of the mechanism is fabricated by using Wire-Electrical-Discharge-Machining (WEDM) process. Testing results indicate that the proposed micromanipulator demonstrates good performance.

This paper focuses on improving tracking performance of totally uncoupled compliant micromanipulator based on robust control method of elliptical hysteresis model. The hysteresis model and hysteresis compensation model for proposed mechanism based on Piezoelectric transducer (PZT) are established by using elliptical model method. The values of elliptical hysteresis model parameters are identified by simulating method with different frequencies of control input. The uncertainty model is also established, the values of corresponding estimated parameters are conformed by experiment method. Based on the uncertainty model and elliptical hysteresis model, the robust tracking control method is presented and utilized to evaluate the tracking performance of proposed mechanism under inputting different tracking curves. The proposed method can accurately control the output displacement and improve tracking performance of this mechanism, which are validated and carried out by using experimental studies. Additionally, the coupling errors between two directions are kept within 0.14% and the tracking errors for different curves are within 2.5%.

The four wing unmanned aerial vehicle owns the characteristics of small size, light weight, convenient operation and well stability. But it is easily disturbed by external environmental factors during flight with these disadvantages of short endurance and poor attitude solving ability. For solving these problems, a microprocessor based on STM32 chip is designed and the overall development is completed by the resources such as built-in timer and multi-function mode general-purpose input/output provided by the master micro controller unit, together with radio receiver, attitude meter, barometer, electronic speed control and other devices. The unmanned aerial vehicle can be remotely controlled and send radio waves to its corresponding receiver, control the analog level change of its corresponding channel pins. The master control chip can analyze and process the data to send multiple sets pulse signals of pulse width modulation to each electronic speed control. Then the electronic speed control will transform different pulse signals into different sizes of current value to drive the motor located in each direction of the frame to generate different rotational speed and generate lift force. To control the body of the unmanned aerial vehicle, so as to achieve the operator's requirements for attitude control, the PID controller based on Kalman filter is used to achieve quick response time and control accuracy. Test results show that the design is feasible.

This article focuses on improving the BiSeNet v2 bilateral branch image segmentation network structure, enhancing its learning ability for spatial details and overall image segmentation accuracy. A modified network called “BiconvNet” is proposed. Firstly, to extract shallow spatial details more effectively, a parallel concatenated strip and dilated (PCSD) convolution module is proposed and used to extract local features and surrounding contextual features in the detail branch. Continuing on, the semantic branch is reconstructed using the lightweight capability of depth separable convolution and high performance of ConvNet, in order to enable more efficient learning of deep advanced semantic features. Finally, fine-tuning is performed on the bilateral guidance aggregation layer of BiSeNet v2, enabling better fusion of the feature maps output by the detail branch and semantic branch. The experimental part discusses the contribution of stripe convolution and different sizes of empty convolution to image segmentation accuracy, and compares them with common convolutions such as Conv2d convolution, CG convolution and CCA convolution. The experiment proves that the PCSD convolution module proposed in this paper has the highest segmentation accuracy in all categories of the Cityscapes dataset compared with common convolutions. BiConvNet achieved a 9.39% accuracy improvement over the BiSeNet v2 network, with only a slight increase of 1.18M in model parameters. A mIoU accuracy of 68.75% was achieved on the validation set. Furthermore, through comparative experiments with commonly used autonomous driving image segmentation algorithms in recent years, BiConvNet demonstrates strong competitive advantages in segmentation accuracy on the Cityscapes and BDD100K datasets.