Leveraging space-based optical platforms for space debris and defunct spacecraft detection presents several advantages, including a wide detection range, immunity to cloud cover, and the ability to maintain continuous surveillance on space targets. As a result, it has become an essential approach for accomplishing tasks related to space situational awareness. However, the prediction of the orbits of space objects is crucial for the success of such missions, and current technologies face challenges related to accuracy, reliability, and practical efficiency. These challenges limit the performance of space-based optical space situational awareness systems. To drive progress in this field and establish a more effective and reliable space situational awareness system based on space optical platforms, this paper conducts a retrospective overview of research advancements in this area. It explores the research landscape of orbit determination methods, encompassing orbit association methods, initial orbit determination methods, and precise orbit determination methods, providing insights from international perspectives. The article concludes by highlighting key research areas, challenges, and future trends in current space situational awareness systems and orbit determination methods.

Accurate detection of non-cooperative spacecraft based on optical sensor data is essential for critical space tasks, such as on-orbit servicing, rendezvous and docking, and debris removal. Traditional object detection methods struggle in the challenging space environment, which includes extreme variations in lighting, occlusions, differences in image scale. To address this problem, this paper proposes a high-precision, deep-learning-based, domain-adaptive detection method specifically tailored for non-cooperative spacecraft. The proposed algorithm focuses on two key elements: dataset creation and network structure design. First, we develop a spacecraft image generation algorithm using CycleGAN, facilitating seamless conversion between synthetic and real spacecraft images to bridge domain differences. Second, we combine a domain-adversarial neural network with YOLOv5 to create a robust detection model based on multi-scale domain adaptation. This approach enhances the YOLOv5 network’s ability to learn domain-invariant features from both synthetic and real spacecraft images. The effectiveness of our high-precision domain-adaptive detection method is verified through extensive experimentation. This method enables several novel and significant space applications, such as space rendezvous and docking and on-orbit servicing.

The successful detection of key components within satellites is a crucial prerequisite for executing on-orbit capture missions. Due to the inherent data-driven functionality, deep learning-based component detection algorithms rely heavily on the scale and quality of the dataset for their accuracy and robustness. Nevertheless, existing satellite image datasets exhibit several deficiencies, such as the lack of satellite motion states, extreme illuminations, or occlusion of critical components, which severely hinder the performance of detection algorithms. In this work, we bridge the gap via the release of a novel dataset tailored for the detection of key components of satellites. Unlike the conventional datasets composed of synthetic images, the proposed Typical Components of Satellites (TYCOS) dataset comprises authentic photos captured in a simulated space environment. It encompasses three types of satellite, three types of key components, three types of illumination, and three types of motion state. Meanwhile, scenarios with occlusion in front of the satellite are also taken into consideration. On the basis of TYCOS, several state-of-the-art detection methods are employed in rigorous experiments followed by a comprehensive analysis, which further enhances the development of space scene perception and satellite safety.

Extracting key features on non-cooperative space targets is a prerequisite for completing primary space missions such as target localization and attitude measurement. However, the space environment is characterized by extreme variations in light intensity, occlusions, and significant differences in image scales, making traditional feature extraction methods challenging to apply to extracting feature points from non-cooperative space vehicles. This paper proposes a deep learning-based keypoints regression algorithm to improve accuracy in non-cooperative space targets. Unlike traditional methods, which struggle with occlusions and lighting changes, our method leverages a hybrid attention mechanism and high-resolution networks to address these challenges effectively. Firstly, a keypoints regression network model for non-cooperative spacecraft is proposed, using a high-resolution network as the backbone. The model incorporates a hybrid attention mechanism during feature extraction and expands the receptive field during feature fusion to improve the accuracy of keypoints regression in occluded scenarios. Secondly, to address high memory usage issues and poor real-time performance caused by complex network models, a lightweight design is applied to the HRNet, ensuring accuracy and model efficiency. Finally, multiple experimental results validate the performance of the proposed deep learning-based keypoints regression algorithm for non-cooperative space targets.

Enabling full robotic workloads with diverse behaviors on mobile systems with stringent resource and energy constraints remains a challenge. In recent years, attempts have been made to deploy single-accelerator-based computing platforms (such as GPU, DSP, or FPGA) to address this challenge, but with little success. The core problem is two-fold: firstly, different robotic tasks require different accelerators, and secondly, managing multiple accelerators simultaneously is overwhelming for developers. In this paper, we propose PIRT, the first robotic runtime framework to efficiently manage dynamic task executions on mobile systems with multiple accelerators as well as on the cloud to achieve better performance and energy savings. With PIRT, we enable a robot to simultaneously perform autonomous navigation with 25 FPS of localization, obstacle detection with 3 FPS, route planning, large map generation, and scene understanding, traveling at a max speed of 5 miles per hour, all within an 11W computing power envelope.