Recent progress in stereo-based 3D Gaussian Splatting (3DGS) SLAM has enabled small-scale robots, which are too small to carry depth cameras, to achieve localization and reconstruct photorealistic scenes with high-speed rendering. However, initializing 3D Gaussians from binocular vision still requires further improvement, and the potential of robot proprioception has not been fully leveraged. This work presents a robust stereo 3DGS SLAM with efficient inertial-legged fusion for small-scale quadruped robots (SaQu-SLAM). We develop a light-weight network to densely initialize the 3D Gaussians in the space. Besides, an efficient fusion method of inertial and legged encoder data based on Kalman filter is introduced. To improve the cross-platform generalization of our algorithm, multiple configuration combinations of these three types of sensors are provided. Moreover, we propose a mode-switching mechanism to handle intermittent visual failures. At last, we perform evaluation on a benchmark dataset, which includes large- and small-scale scenes, and a small quadruped robot in real-world confined-scale scenes, reducing the absolute trajectory error by an average of 19%, 13% and 25% respectively, when compared with other state-of-the-art methods in a similar context. It is also the only successful method in our self-customized confined mixed textured and textureless scene, whereas all vision-based or visual-inertial methods fail. Our system achieves real-time performance even on an embedded platform (Jetson AGX Orin).

This paper presents a robust visual SLAM system designed for small-scale quadruped robots (ViQu-SLAM) for accurate localization, especially to mitigate the issue of erroneous data association caused by moving objects in dynamic environments. The proposed approach leverages a selfadaptive framework that integrates semantic segmentation with alterations in the spatial location of categorized map points. Besides, combination of leg odometry derived from forward kinematics with IMU provides scale information for positional transformations between keyframes, thus optimizing the overall localization accuracy of quadruped robots. At last, we performed evaluation across various stages and the results demonstrate competitive performance, with 53.16% reduction in average absolute trajectory error compared to that of ORB-SLAM3 in dynamic benchmark datasets. As a result, ViQu-SLAM, including visual and IMU-fused leg odometry, exhibits promising results on a small quadruped robot, reducing positioning errors in dynamic scenes by an average of 29.36% compared to existing state-of-the-art methods.

For most machine learning models, the mapping from the hyper-parameter set to the model's generalization error can be regarded as a complex black box function. Particle swarm optimization (PSO) methods cannot be directly used in the problem of hyper-parameters estimation since the mathematical formulation of the mapping from hyper-parameters to loss function or generalization accuracy is unclear. Functions with high evaluation costs can be solved by Bayesian optimization (BO) which converting the optimization of hyper-parameters into the optimization of an acquisition function. The proposed method in this paper uses the particle swarm method to optimize the acquisition function in the BO to get better hyper-parameters. The performances of proposed method in both of the classification and regression models are evaluated and demonstrated.

The research of CAPTCHA recognition is helpful to discover the security vulnerabilities in time and improve its safety. In comparison with digits and English letters, Chinese characters have many more categories which lead to the requirement of a large amount of training data. Therefore, this study proposes a novel method for one‐shot and few‐shot Chinese CAPTCHA recognition, using the deep Siamese network, based on the idea of template matching. In this method, the residual convolutional neural network branches are used for feature extraction of CAPTCHAs, a fully‐connected layer is used for calculating the similarity of features, and a hard negative mining algorithm is designed to promote convergence. Experiments are done on a self‐built small‐scale Chinese CAPTCHA dataset. The results show that this proposed method can achieve higher accuracy on the known characters than traditional methods. For the brand‐new characters, only one template is required to recognise them and the accuracy is close to known characters. To summarise, it is able to build a Chinese CAPTCHA recognition model with high accuracy and extensibility by using a small‐scale dataset.

The knowledgebase uncertainty and the argument preferences are considered in this paper. The uncertainty is captured by weighted satisfiability degree, while a preference relation over arguments is derived by the beliefs of an agent.

Algorithm's time complexity is an essential issue for time series prediction in numerous practices.A novel fast exact inference method for Gaussian process model is proposed in this paper to accelerate the task of non-stationary time series prediction. Experiment was done on the real world power load data.

Inferring the causal direction between two variables is a nontrivial problem in the subject of causal discovery from observed data. A method for errors-in-variables models where both the cause variable and the effect variable are observed with measurement errors is presented in this paper.