Foundation Models Enable Autonomous Collision Avoidance in Congested Orbital Environments

The growing density of space objects in low Earth orbit (LEO), driven by the deployment of large satellite constellations, has elevated the risk of orbital collisions and the need for high-precision conjunction analysis. Traditional methods based on Two-Line Element (TLE) data suffer from limited accuracy and insufficient uncertainty modeling. This study proposes a hybrid collision assessment framework that combines Monte Carlo simulation, spline-based refinement of the time of closest approach (TCA), and a multi-stage deterministic refinement process. The methodology begins with probabilistic sampling of TLE uncertainties, followed by a coarse search for TCA using the SGP4 propagator. A cubic spline interpolation then enhances temporal resolution, and a hierarchical multi-stage refinement computes the final TCA and minimum distance with sub-second and sub-kilometer accuracy. The framework was validated using real-world TLE data from over 2600 debris objects and active satellites. Results demonstrated a reduction in average TCA error to 0.081 s and distance estimation error to 0.688 km. The approach is computationally efficient, with average processing times below one minute per conjunction event using standard hardware. Its compatibility with operational space situational awareness (SSA) systems and scalability for high-volume screening make it suitable for integration into real-time space traffic management workflows.

Characterizing Resident Space Objects (RSOs) has become paramount for several Space Situational Awareness functions, such as accurate orbit prediction, collision avoidance and sensor tasking. Due to the huge and diverse amount of data to handle and fuse for this purpose, there is an increasing interest in Machine Learning (ML) approaches to retrieve physical parameters of RSOs in a cost-efficient manner. Among different ML architectures, Recurrent Neural Networks (RNNs) are particularly suitable to handle sequential or time-series data. In this context, this paper demonstrates the possibility to use Recurrent Neural Networks to estimate the ballistic coefficient of RSOs in Low Earth Orbit from time series of orbital elements. A particular RNN architecture has been adopted, i.e., the Gated Recurrent Unit, as it has been demonstrated to be cost efficient and to have good performance for the required application. The sensitivity of RNNs with respect to atmospheric model uncertainty has been analysed. The algorithm can handle unevenly spaced time-series data. The proposed neural network has been demonstrated to be robust with respect to orbital state errors. The applicability of the presented approach is tested and discussed using synthetic datasets generated with a high-accuracy numerical orbital propagator, reaching a mean percentage error of 10% in the estimation of the ballistic coefficient.

As the number of man‐made Earth‐orbiting objects increases, satellite operators need enhanced space traffic management capabilities to ensure safe space operations. For objects in Low Earth orbit, orbit determination and prediction require accurate estimates of the local thermospheric density. In previous work, the estimation of thermospheric densities using two‐line element data and a reduced‐order model for the upper atmosphere was demonstrated. In this study we demonstrate an approach for density estimation using radar and GPS tracking data. For this, we assimilate the tracking data in a dynamic reduced‐order density model using a Kalman filter by simultaneously estimating the orbits and global density. We used the radar range and range rate measurements of 20 objects and the GPS position measurements of 10 commercial satellites. The estimated density was validated against accurate SWARM density data and compared with NRLMSISE‐00, JB2008, and two‐line element (TLE)‐estimated densities. We found that the estimated densities are significantly more accurate than NRLMSISE‐00 and JB2008 densities. In particular, using the GPS data of 10 satellites, we obtain density estimates with a daily 1‐σ error of only 5% compared to 14% and 22% for empirical models and 10% for TLE‐estimated density. These accurate density estimates can be used to improve orbit determination and the use of real‐time tracking data would enable real‐time density estimation.

The Changchun Observatory of the National Astronomical Observatories, Chinese Academy of Sciences, and the Shanghai Astronomical Observatory are used to generate very short arc (VSA) angle observations of objects in low Earth orbit (LEO) and geostationary orbit (GEO) with their ground-based electrical–optical telescope arrays (EA), the Changchun EA and SAO FocusGEO, respectively. These observations are used in this paper. The range-searching (RS) algorithm for initial orbit determination (IOD) is improved through the multiple combinations of observations and the dynamic range-searching step length. Two different computation modes (the normal mode and the refining mode) of the IOD computation process are proposed. The geometrical method for the association is used. The IOD and association methods are extended to the real optical observations for both LEO and GEO objects. The results show that the IOD success rate of arcs from the LEO objects is about 91%, the error of the semimajor axis (SMA) of the initial orbital elements is less than 50 km, and the correlation accuracy rate is about 89%. The IOD success rate of arcs from the GEO objects is higher than 88%, and the correlation accuracy rate is greater than 87%. The recent COSMOS 1408 antisatellite test (ASAT) generated a large amount of debris. The algorithm of this paper and the observations of Changchun EA are used to initially identify new debris, possibly from the ASAT through initial orbit determination and track association. Finally, 64 suspected new pieces of debris can be found. The results show the effectiveness of the IOD and the correlation algorithm, as well as the potential application of the optical–electrical array in studying space events.

Active debris multi-removal mission concept based on hybrid propulsion

Track detection of high-velocity resident space objects in Low Earth Orbit

With escalating congestion in the electromagnetic spectrum, commensal (passive) sensors provide a solution for detecting objects of interest without the need for bespoke transmitters. Through the characterization and spatial assessment of existing broadcast infrastructure, illuminators of opportunity constitute an abundance of transmitters and allow for wide-area commensal sensor coverage of objects in space. This paper describes work to evaluate the potential of a FM broadcast based commensal sensor to detect space objects in Low Earth Orbit (LEO). Specifically, a planning tool has been developed to assess and simulate passes of objects such as the International Space Station (ISS). With reference to a Mission Planning Tool (MPT), we describe some of the important system considerations for using commensal sensors as a ground-based sensor for LEO space observation.

Conjunction assessment of space objects in Low Earth Orbit (LEO) generally uses information collected by ground-based space surveillance sensors. These sensors track both the primary object (normally an active satellite) and the secondary object (typically space debris). The tracking data is used to update both objects’ orbits for collision risk assessment. The primary satellite’s involvement in this process is that of a satellite in jeopardy - the primary satellite does not usually contribute tracking data on the secondary as they are typically unequipped to do so. In this paper, an examination how an at-risk LEO primary satellite could obtain optical tracking data on a secondary object prior to the Time of Closest Approach (TCA) and assess its own collision risk without the need for additional ground-based space surveillance data is performed. This analysis was made possible by using in-situ optical measurements of space objects conjuncting with the Canadian NEOSSat Space Situational Awareness R&D microsatellite. By taking advantage of the near “constant-bearing, decreasing range” observing geometry formed during a LEO conjunction, NEOSSat can collect astrometric and photometric measurements of the secondary object in the time prior to TCA, or in the multiple half-orbits preceding TCA. This paper begins by describing the in-situ phenomenology of optically observed conjunctions in terms of the observing approach, geometry and detected astrometric and photometric characteristics. It was found that conjuncting objects are detectable to magnitude 16 and astrometric observations can be used for position covariances in the computation of probability of collision. Illustrative examples are provided. In orbits prior to TCA, in-track positioning error is improved by a factor of two or more by processing space-based observations on a filtered position estimate of the secondary. However, cross-track positioning knowledge is negligibly improved due to the inherent astrometric measurement precision of the NEOSSat sensor and the oblique observing geometry during conjunction observations. A short analysis of object detectability where star trackers could be used to perform similar observations finds that larger payload-sized objects would generally be detectable. However, smaller debris objects would require higher sensitivity from the star tracker if employed for optical conjunction derisk observations.

Atmospheric mass density (AMD) plays a vital role in the drag calculation for space objects in low-Earth orbit. Many empirical AMD models have been developed and used for orbit prediction (OP) and efforts continue to improve their accuracy in forecasting high-altitude atmospheric conditions. Previous studies have assessed these models at the height of 200 km to 600 km. In this article, four state-of-the-art AMD models, i.e., Mass Spectrometer Incoherent Scatter extended model (MSISE90), MSISE00, Jacchia–Bowman 2008, and Drag Temperature Model 2013 are assessed for their OP capabilities by using a new data source of Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) satellite ephemerides at an orbital height of $\sim$800 km, where the contribution of ions in the total AMD is more significant. A new testing model was developed by accounting for ion contribution based on the International Reference Ionosphere 2016 model, including many more ion species that are not accounted for in other AMD models. In the assessment, two periods of forty days were chosen in 2014–2015 and 2018–2019, representing solar maximum and minimum periods, respectively, to assess four existing AMD models and the proposed model. Thorough analyses were conducted to compare OP results using different AMD models with precise reference ephemerides of COSMIC satellites and based on various space weather indices. It is shown that the proposed model outperforms all other AMD models in terms of OP errors during the solar maximum period. During solar minimum, the drag acceleration is relatively small for COSMIC satellites. Assessment of all AMD models in the OP process tends to be contaminated by the remaining uncertainty sources, such as solar radiation pressure.

Accuracy of Two-Line-Element Data forGeostationary and High-Eccentricity Orbits

In this paper we consider the use of wide field of view radar sensors such as the Murchison Widefield Array (MWA), a low frequency radio telescope designed for astrophysics and cosmology, for rapid response observations of the debris clouds produced by collisions between objects in Earth orbit. With an increasing density of objects in Low Earth Orbit, including legacy assets used by the astronomy community over decades, the risk of new debris clouds forming is also increasing. The MWA constitutes a wide field, rapid response passive radar system and we explore its likely performance in the detection and characterisation of debris clouds. In general, astronomy facilities such as the MWA can play a role in protecting the space environment for the future. In order to undertake this work, we adapt the NASA EVOLVE 4.0 breakup model, utilising the EVOLVE outputs to produce representative dynamic debris clouds. We find that the MWA is likely to detect a large fraction (>70%) of modelled debris cloud fragments for collision masses between 100 kg and 1000 kg for orbits in the lower part of LEO, if the MWA can achieve close to optimal detection sensitivity. Useful detection fractions are still achieved for more conservative assumptions. The detection fraction of fragments decreases as a function of altitude and inversely with collision mass. Encouragingly, we find that the wide field nature of the MWA allows the full evolving debris clouds to be observed in a single observation, with only $\sim2\%$ of the debris fragments escaping the sensitive portion of the field of view after 100 seconds, for all collision masses and altitudes. These results show that the MWA is an intrinsically useful facility for the rapid characterisation of debris clouds, but that work is required to achieve the data processing within an appropriate timeframe to provide rapid alerts.

The continuously growing number of objects orbiting around the Earth is expected to be accompanied by an increasing frequency of objects re-entering the Earth’s atmosphere. Many of these re-entries will be uncontrolled, making their prediction challenging and subject to several uncertainties. Traditionally, re-entry predictions are based on the propagation of the object’s dynamics using state-of-the-art modelling techniques for the forces acting on the object. However, modelling errors, particularly related to the prediction of atmospheric drag, may result in poor prediction accuracies. In this context, we explored the possibility of performing a paradigm shift, from a physics-based approach to a data-driven approach. To this aim, we present the development of a deep learning model for the re-entry prediction of uncontrolled objects in Low Earth Orbit (LEO). The model is based on a modified version of the Sequence-to-Sequence architecture and is trained on the average altitude profile as derived from a set of Two-Line Element (TLE) data of over 400 bodies. The novelty of the work consists in introducing in the deep learning model, alongside the average altitude, and three new input features: a drag-like coefficient (B*), the average solar index, and the area-to-mass ratio of the object. The developed model was tested on a set of objects studied in the Inter-Agency Space Debris Coordination Committee (IADC) campaigns. The results show that the best performances are obtained on bodies characterised by the same drag-like coefficient and eccentricity distribution as the training set.

Simulation modelling of the observability of low Earth orbit (LEO) objects was performed using optical surveillance facilities depending on their geographic location and time of year. Orbital data for LEO objects from the open-access catalogue of the near-Earth space objects of the US Combined Space Operations Center (CSpOC) were taken as the initial data for the simulation. The simulation results revealed a complex relationship between the pattern of observability of a LEO object, its orbital parameters and location of the optical surveillance facility, in particular, for Sun-synchronous orbits (SSO) and observing facilities located near the equator. We also discuss variations in the frequency of passes of LEO objects into the field of view (FOV) and in the duration of their observation while passing through the FOV for optical surveillance facilities at three alternative locations. The obtained results and modelling techniques can be further used in the location planning of new optical observing facilities.

There is increasing interest in space domain awareness worldwide, motivating investigation of the use of non-traditional sensors for space surveillance. One such class of sensor is VHF wind profiling radars, which have a low cost relative to other radars typically applied to this task. These radars are ubiquitous throughout the world and may potentially offer complementary space surveillance capabilities to the Space Surveillance Network. This paper updates an initial investigation on the use of Buckland Park VHF wind profiling radars for observing resident space objects in low Earth orbit to further investigate the space surveillance capabilities of the sensor class. The radar was operated during the Australian Defence “SpaceFest” 2019 activity, incorporating new beam scheduling and signal processing functionality that extend upon the capabilities described in the initial investigation. The beam scheduling capability used two-line element propagations to determine the appropriate beam direction to use to observe transiting satellites. The signal processing capabilities used a technique based on the Keystone transform to correct for range migration, allowing the development of new signal processing modes that allow the coherent integration time to be increased to improve the SNR of the observed targets, thereby increasing the detection rate. The results reveal that 5874 objects were detected over 10 days, with 2202 unique objects detected, representing a three-fold increase in detection rate over previous single-beam direction observations. The maximum detection height was 2975.4 km, indicating a capability to detect objects in medium Earth orbit. A minimum detectable RCS at 1000 km of −10.97 dBm2 (0.09 m2) was observed. The effects of Faraday rotation resulting from the use of linearly polarised antennae are demonstrated. The radar’s utility for providing total electron content (TEC) measurements is investigated using a high-range resolution mode and high-precision ephemeris data. The short-term Fourier transform is applied to demonstrate the radar’s ability to investigate satellite rotation characteristics and monitor ionospheric plasma waves and instabilities.

The rising population of artificial satellites and associated debris in low-altitude orbits is increasing the overall brightness of the night sky, threatening ground-based astronomy as well as a diversity of stakeholders and ecosystems reliant on dark skies. We present calculations of the potentially large rise in global sky brightness from space objects in low Earth orbit, including qualitative and quantitative assessments of how professional astronomy may be affected. Debris proliferation is of special concern: we calculate that all log-decades in debris size contribute approximately the same amount of night sky radiance, so debris-generating events are expected to lead to a rapid rise in night sky brightness along with serious collision risks for satellites from centimetre-sized objects. This increase in low-Earth-orbit traffic will lead to loss of astronomical data and diminish opportunities for ground-based discoveries as faint astrophysical signals become increasingly lost in the noise. Lastly, we discuss the broader consequences of brighter skies for a range of sky constituencies, equity/inclusion and accessibility for Earth- and space-based science, and cultural sky traditions. Space and dark skies represent an intangible heritage that deserves intentional preservation and safeguarding for future generations. Each space launch is assessed for various risks, but not its wider impacts. This Perspective shows how the aggregate effects of space launches, plus the attendant rise of space debris, affect the darkness of our night sky now and in the future.