OPS-SAT is a 3U CubeSat launched on December 18, 2019, it is the first nanosatellite to be directly owned and operated by the European Space Agency (ESA). The spacecraft is a flying platform that is easily accessible to European industry, institutions, and individuals for rapid prototyping, testing, and validation of their software and firmware experiments in space at no cost and no bureaucracy. Equipped with a full set of sensors and actuators, it is conceived to break the “has never flown, will never fly” cycle. OPS-SAT has spearheaded many firsts with in-orbit applications of Artificial Intelligence (AI) for autonomous operations. AI is of rising interest for space-segment applications despite limited technology demonstrators on-board flying spacecrafts. Past missions have restricted AI to inferring models trained on the ground prior to being uplinked to a spacecraft. This paper presents how the OPS-SAT Flight Control Team (FCT) breaks away from this trend with various AI solutions for in-flight autonomy. Three on-board case studies are presented: 1) image classification with Convolutional Neural Network (CNN) model inferences using TensorFlow Lite, 2) image clustering with unsupervised learning using k-means, and 3) supervised learning to train a Fault Detection, Isolation, and Recovery (FDIR) model using online machine learning algorithms. CNN inference with TensorFlow Lite running on-board the spacecraft showcases in-space application of an industry standard open-source solution originally developed for terrestrial edge and mobile computing. Furthermore, the solution is “openable” with an inference pipeline that can be constructed from crowdsourced trained models. This mechanism enables open innovation methods to extend on-board ML beyond its original mission requirement while stimulating knowledge transfer from established AI communities into space applications. Further classification is achieved by re-using an open-source k-means algorithm to cluster images in groups of “cloudiness” and initial results in image segmentation (feature extraction) have promising outlooks. Results from training a FDIR model to protect the spacecraft's camera lens against direct exposure to sunlight are presented, achieving balanced accuracies ranging from 85% to 99% from models trained with the Adagarad RDA, AROW, and NHERD online ML algorithms in multi-dimensional input spaces using photodiode diagnostics data stream as training data. The ability to train models in-flight with data generated on-board without human involvement is an exciting first that stimulates a significant rethink on how future missions can be designed.

Cost-effective techniques for systematic archaeological prospection are essential to improve the efficiency of preventive archaeology and the preservation of cultural heritage. Web Mapping Services, such as Microsoft Bing Maps, provide imagery covering extensive areas at high resolution. These can, in some cases, reveal cropmarks of buried historical structures. Given that archaeological prospection is not generally the priority of most common Web Mapping Services, the conditions under which images are acquired are not always suitable for the appearance of cropmarks. Therefore, their detection is typically serendipitous. This pilot project attempts to assess the potential to use the Microsoft Bing Maps Bird’s Eye service within a crowdsourcing platform to systematically search for archaeological cropmarks in the surroundings of the city of Rome in Italy. On this platform, which is hosted by the company Scifabric (Southampton, UK) and based on PyBossa, an Open Source framework for crowdsourcing, members of the public are invited to interpret oblique air photo tiles of Bing Maps Bird’s Eye. While the project is still on-going, at least one seamless coverage of tiles in the area of interest has been interpreted. For each tile, the Bing Maps Bird’s Eye service provides oblique air photo coverage in up to four possible orientations. As of 5 July 2020, 18,765 of the total 67,014 tasks have been completed. Amongst these completed tasks, positive detections of cropmarks were recorded once for 1447 tasks, twice for 57 tasks, and three or more times for 10 tasks. While many of these detections may be erroneous, some correspond with archaeological cropmarks of buried remains of buildings, roads, aqueducts, and urban areas from the Roman period, as verified by comparison with archaeological survey data. This leads to the conclusion that the Bing Maps Bird’s Eye service contains a wealth of information useful for archaeological prospection, and that to a certain extent citizen researchers could help to mine this information. However, a more thorough analysis would need to be carried out on possible false negatives and biases related to the varying ease of interpretation of residues of different archaeological structures from multiple historical periods. This activity forms the first part of a research project on the systematic prospection of archaeological cropmarks. The ultimate aim is to reach a critical mass of training data through crowdsourcing which can be augmented and used as input to train a machine learning algorithm for automatic detection on a larger scale.

The open-source software movement has transformed the way software is developed. It has leveraged the power of software reuse with sustainable community support to attractively develop and scale software for the benefit of all. The movement has found its way into the space domain where newer players are stepping in and adopting open-source solutions for both ground and space segments. A notable adoption driver are the recent developments in off-the-shelf compute capabilities of modern spacecraft coupled with the increasing decision-making complexity and data-generation capacity of modern satellite payloads. The latter requires ever more on-board autonomy due to communication bandwidth limitations with the ground control software and the inherent human factor limitations in decision-making. Software which was never expected or even intended to be used in the space domain is now being adopted by space players, particularly in New Space. In support of industry and academia to explore and experiment with open-source for space operations, the European Space Agency (ESA) launched the OPS-SAT Space Lab in December 2019. The spacecraft is an open innovation platform in the form of a 3U CubeSat in Low Earth Orbit (LEO). OPS-SAT has accomplished many firsts in space powered by open-source and rapid prototyping, from the first securities trade to pioneering the use of Artificial Intelligence (AI) frameworks for Machine Learning (ML) training and inferences with payload and telemetry data. This paper presents lessons learned and a suggested way forward from the perspective of two former OPS-SAT Flight Control Team (FCT) members turned experimenters, showcasing the opportunities and challenges in revisiting how spacecraft operations benefit from a rethink with open-source software as a pillar of mission design.

View Video Presentation: https://doi.org/10.2514/6.2022-2508.vid The SmartCam software on the OPS-SAT spacecraft is the first use of Artificial Intelligence (AI) by the European Space Agency (ESA) for autonomous planning and scheduling on-board a flying mission. The software’s geospatial capability autonomously triggers image acquisitions when the spacecraft is above areas of interest. Inferences from an on-board Machine Learning (ML) model classifies the captured pictures for downlink prioritization. This is made possible by the spacecraft’s powerful processors, capable of running open-source software originally developed for terrestrial systems. Notably, with the GEOS Geometry Engine for geospatial computations, the TensorFlow Lite framework for ML model inferences, and a k-means algorithm for image clustering. These features provide new perspectives on how space operations can be designed for future missions given greater in-orbit compute capabilities.

The camera onboard the European Space Agency’s OPS-SAT-1 spacecraft has been operating past its design life. As the payload ages, there is an increasing risk of sensor degradation leading to corrupted images. This paper evaluates Generative AI models with Wasserstein GANs (WGANs)—an enhanced type of Generative Adversarial Networks—as a noise reduction solution to reconstruct noisy images directly onboard the spacecraft. Autoencoder neural networks are also trained and evaluated for comparative purposes given their common use in noise reduction. Images downlinked from the spacecraft serve as training data to which artificial fixed-pattern noise is applied to simulate sensor degradation. The trained neural networks are uplinked to the spacecraft’s edge computer payload where they are processed by the onboard TensorFlow Lite interpreter to output the reconstructed images. On September 29, 2023, the OPS-SAT-1 mission achieved a significant milestone when it successfully captured, noised, and subsequently denoised two images using WGANs, marking the pioneering first application of Generative AI in space. The restored images have remarkably high structural similarity indices of 0.894 and 0.922—where 1 would indicate that they are identical to their original images. Interestingly, some reconstructed images are more confidently labeled by the onboard convolutional neural network image classifier than their original counterparts. The counterintuitive observation challenges the conventional understanding that higher resolution always yields better results. This suggests that simplifying or modifying certain data features enhances the ability of some models to accurately interpret given inputs.

The European Space Agency (ESA)’s OPS-SAT-1 mission pioneered the application of edge computing for onboard flight operations. Its objective is to upload and test new cutting edge on-board software in space on a daily basis. This involves a much more intense use of the on-board control computer compared to traditional missions. After several years, radiation damage occurred that would have normally led to mission end. The mission control team was able to restore the core payload functionality by exploiting the reconfigurability of the computing platform. The Satellite’s Experimental Processing Platform (SEPP) is equipped with an 800MHz processor and a reconfigurable Field Programmable Gate Array (FPGA). The SEPP is not radiation-hardened and runs embedded Linux. Since the beginning of 2022, several instances of file corruption in the SEPP’s non-volatile embedded MultiMediaCard (eMMC) memory, physically a NAND flash, were observed. Earlier instances could be rectified with conceptually simple procedures like rebuilding the filesystem. However, an event in June 2022 resulted in a total loss of communication with the payload. A previously unused access route via UART had to be made active before the problem could be investigated. The team then narrowed down the origin of the problem and revealed that the eMMC was not totally lost but that writing to it could crash the controller. Novel ways of avoiding memory writes had to be found to continue the mission. This approach ushered in an entirely new fully volatile mode of SEPP operations. At the same time a recovery strategy, in case the eMMC memory completely failed, also had to be designed and implemented. This entailed storage optimisation and added redundancy. This paper delves into the novel eMMC defect and sheds light on the resourceful strategies that were employed for its recovery. It outlines various approaches and provides a comparative analysis to underscore the associated trade-offs. The SEPP was custom built for OPSSAT-1 at a time when commercial off-the-shelf (COTS) counterparts did not exist. Since then, similar payloads have become readily available COTS products and their adoption in small satellite missions has sky-rocketed. This paper aims to share the Mission Control Team (MCT)’s experiences with managing and recovering from radiation induced memory faults in spacebound edge computers and propose guidelines for future small satellite missions carrying similar equipment.