Harmonic drive systems are widely used in joints of space manipulators because of their low relative quality, compact structure and low costs. However, dynamic characteristics, including kinematic error, stiffness, and friction of harmonic drive systems, hinder the performance improvements of space manipulators, such as pointing accuracy and stability. To overcome this problem, scholars research the experiments, analyses and models on the dynamic characteristics of harmonic drive systems. This paper reviews studies about kinematic error, stiffness and friction. Some research topics are presented.

Low altitude airspace drones are characterized by small size and flexible flight, which brings difficulties to visual detection of trespassing drones. The low altitude drone object segmentation algorithm incorporating an attention mechanism named Rep YOLACT (re-parameterization you only look at coefficients network) is proposed, which is first used with RepVGG (re-parameterization visual geometry group) networks to improve ResNet (residual network) backbone in YOLACT and enhance the feature extraction capability of the network. Meanwhile, CBAM (convolutional block attention module) is added after the three feature layers output from the backbone feature extraction network, so as to further utilize the information of the feature layers efficiently. Experiments are conducted on FL-drones (flying drones dataset) and MUD (multiscale unmanned aerial vehicle dataset), respectively. The results show that the proposed Rep YOLACT algorithm improves mask AP (average precision) and mask AR (average recall) by 0.3% and 11.7%, respectively, compared with YOLACT algorithm on FL-drones. The proposed Rep YOLACT algorithm improves 2.3% and 5% on mask AP and prediction frame AR compared to YOLACT algorithm, which can perform the drone segmentation task well and its segmentation accuracy is higher than other mainstream segmentation algorithms.

Aiming at the problem of poor robustness of the current spatial noncooperative CubeSat pose estimation algorithm using image features, a design scheme using CubeSat vertices for pose estimation method is proposed in this paper. Based on the grayscale image obtained by binocular vision, HED (holistically nested edge detection) network and binary morphology processing method are used to improve the robustness of edge extraction. After detecting the polygon features in the edge image, the polygons with repeated and interference are filtered out, and the key vertices of the CubeSat are identified by a common side double frame key vertices discrimination algorithm, which realizes the estimation of the structural parameters and pose of the CubeSat. The CubeSat model is used for experimental verification. Compared with ICP (iterative closest point) precise registration method, the overall method achieves a maximum deviation of 4.4° and 1.2cm within the detection distance of 30~70cm. In this paper, the edge extraction method improves the discriminant accuracy of target structural parameters by 10%~40%, which provides a new idea for the estimation of structural parameters and pose of noncooperative target CubeSat.

To solve the control problem of nonminimum phase underactuated vertical taking off and landing (VTOL) aircraft, an ACPID(autocoupling proportional integral differrential) control method is proposed. Firstly, coordinate transformation is used to map the center of mass of VTOL aircraft to Huygens vibration center, which can not only realize decoupling of control input of the new system, but also avoid zero dynamic instability of nonminimum phase VTOL aircraft. Then, Huygens vibration center is designed with ACPID controller in vertical and horizontal directions respectively, and the bottom thrust and the virtual instruction of roll attitude angle for VTOL aircraft are obtained respectively, and then the ACPID controller of the roll attitude angle is designed to form the roll torque, so as to realize the position tracking control of VTOL aircraft system. Finally, the robust stability and antidisturbance robustness of the closed loop control system are proved by the complex frequency domain analysis theory. Theoretical analysis and simulation results show the effectiveness of the proposed method, which has important scientific significance and wide application prospects in the field of nonminimum phase underactuated control system.

Combining the characteristics of redundant configuration of attitude sensors and unknown inertia parameters of the on orbit assembled satellites, in order to suppress the influence of sensor measurement errors and parameter uncertainty disturbance on satellite attitude, firstly a gyroscope/GPS/star sensor multisensor federal filtering algorithm is established and the residual chi square test method is used to achieve the detection, isolation and reconstruction of sensor faults, by which the attitude determination precision and result reliability are improved. On this basis, the parameter identification algorithm is used to estimate the assembly satellite inertia parameters, and the adaptive fixed time sliding mode control algorithm is proposed to improve the control precision of the sliding mode control under unknown disturbance and reduce the parameter selection range to attenuate the chattering phenomenon. Finally, the algorithm effectiveness is verified by numerical simulation, and the attitude precision and attitude stability of the assembled satellite are within 0.000 2° and 0.000 3(°)/s, respectively, indicating that the proposed algorithm can achieve high precision and high stability attitude control of the assembled satellites.

This paper proposes a dynamic scene visual odometry method based on deep learning. The C3Ghost module is built using the lightweight Ghost module combined with the target detection network YOLOv5s, and the CA (coordinate attention mechanism) is introduced to improve the network detection speed while ensuring detection accuracy. It is combined with the motion consistency algorithm to eliminate dynamic feature points and only use static feature points for pose estimation. Experimental results show that compared with the traditional ORB SLAM3 (orient FAST and rotated BRIEF simultaneous localization and mapping 3) algorithm, the ATE (absolute trajectory error) and RPE (relative pose error) on the TUM (technical university of Munich) RGB-D (RGB depth) high dynamic data set has improved by more than 90% on average. Compared with the advanced SLAM algorithm, it is also relatively improved. Therefore, this algorithm effectively improves the stability and robustness of visual SLAM in dynamic environments.

Single pixel imaging is an imaging technique that reconstructs a complete image using only nonresolving bucket detectors combined with spatial light modulation information. It features nonlocal imaging and high sensitivity, making it suitable for ultra long distance imaging and detection of noncooperative targets in outer space. However, it requires multiple spatial light modulations for detection, resulting in low signal to noise ratio in the reconstructed images. A global attention mechanism based image enhancement method for low sampling rates is presented in this paper. A novel SUNet(swin transformer unet)network is built via Transformer architecture to address the issues of translational invariance and limited global receptive field in traditional convolutional neural networks. Improved differential ghost imaging algorithm based on CC(cake cutting)sequence is employed to reconstruct low quality images under low sampling conditions, which are then enhanced by SUNet. Experimental results show that, compared to the GIDC(ghost imaging using deep neural network constraint)method proposed in 2022, this approach achieves 3.29 dB improvement in peak signaltonoise ratio and 8% increase in structural similarity at 0.1 sampling rate, providing a new technological avenue for spatial detection in singlepixel imaging.

A high performance anomaly detection model has been constructed to address the problem of sparse anomalous image data in the real world. A two stage framework anomaly detection model is built using only normal training data and a small amount of synthetic anomaly sample. First， a ResNet 18 encoder model is trained to extract representation by the pretext of classifying normal data and synthetic anomaly data. Then， a single classifier for anomaly images is built through modelling the distribution of normal data representations using Gaussian density estimation. GradCAM is applied to extend the model， enabling the anomaly detection model to locate anomaly regions without labels. Finally， experiments are conducted on a simulated anomaly detection dataset using real world images， demonstrating that the proposed algorithm can detect anomaly and provide location results in remote sensing images that are even difficult to recognize with human eyes.

Efficient and accurate anomaly detection of power equipment is essential for aerospace safety. Scientific detection and maintenance can promptly identify potential faults and ensure the safety and reliability of the system. The data collected by sensors from power equipment contains valuable information. Feature extraction is usually required for processing these data. Although deep learning methods historically obtain excellent results, there is always a trade off between fine tuning existing networks or designing models from scratch for sensor data processing. To address this issue, we propose a temporal feature extraction network for time series data based on self supervised learning. First, we use self supervised learning methods to pre train the network. Then we devise a novel network model structure that can effectively extract the representation of time series data. Finally, we evaluate the proposed method on relevant datasets, and the experimental results demonstrate the effectiveness of the proposed method.

This paper investigates the multi-sensor allocation strategy problem for linear system identification in a networked environment. A binary quantization mechanism is introduced to save communication bandwidth. For addressing the redundancy and channel congestion issues during data transmission, a differential event driven communication mechanism is proposed, and its communication rate is provided. It is shown that such mechanism has the capability of complete information recovery. Based on the available data at the receiving end, identification algorithms for various finite impulse response systems are constructed, and their convergence performance is analyzed. Furthermore, the multi-sensor allocation problem under constrained communication resources is modeled as a constrained optimization problem, and an improved genetic algorithm is designed to give the optimal solution. Finally, a numerical example is used to verify the correctness and effectiveness of theoretical results.

In order to ensure safe and accurate landing on celestial bodies, it is necessary to measure the speed of space probe, providing important information for guidance, navigation and control system. A real time visual velocity measurement method only based on optical camera is proposed in this paper. For terrain image sequences of celestial bodies, we use recurrent all pairs field transforms to extract the optical flow field between adjacent frames. Then we extract the eigenvectors corresponding to the optical flow field via the convolution layer and pooling layer in deep neural networks. To reduce the influence on measurement accuracy caused by visual perspective during the landing process, a long short term memory network for video sequences is constructed to match the eigenvectors up with velocities, thus a real time landing speed estimation is achieved for space probes. Simulation results demonstrate that our technique decreases the mean absolute percentage error by 11.98% and has higher measurement accuracy in comparison to the forward propagation network.