Periodic Reporting for period 1 - FAAST (Facilitating Autonomy in Astrodynamics for Spacecraft Technology)
Periodo di rendicontazione: 2023-07-16 al 2025-07-15
Recent decades have seen reduced cost to deliver payloads to orbit, as well as miniaturization and modernization of spacecraft technology. These trends foresee a near future where small, low-cost space missions will be carried out on a regular basis by a variety of public and private organizations. Another trend, mostly visible in terrestrial technologies, is automation, powered in parallel by advances in computing and the proliferation of computers in commercial products.
An important bottleneck in the current state-of-the-art for spaceflight is that the onboard computational capabilities in spacecraft are still quite limited, due to the need for radiation-hardened onboard computers to survive the harsh deep space environment. As a result, deep-space spacecraft must still be cautiously controlled remotely from the ground by teams of engineers, using dedicated deep-space communications infrastructure of limited capacity. Thus, despite the falling cost-to-orbit and the advancement of space technology, operation costs stemming from such activities remain stubbornly high. To propel a scalable future for space exploration, there is a strong need for autonomous capability that can operate with limited computational resources and at a level of safety and reliability worthy for use on invaluable spacecraft.
The main goal of FAAST is to develop robust and computationally efficient algorithms for autonomous orbit guidance and control for low-cost space vehicles which must plan and re-plan their trajectories in an uncertain environment. This is achieved by the following four objectives:
1. Develop and validate a new approach that facilitates computationally feasible onboard techniques.
2. Ensure that the new formulation can be applied to the highly uncertain space environment.
3. Apply the new implementation to develop an algorithm for autonomous orbit guidance in a relevant test problem. A successful algorithm should be computationally lean and robust, facilitating on-board planning and re-planning.
4. High-fidelity testing of the algorithm and study of high-fidelity modeling of the space environment.
An important bottleneck in the current state-of-the-art for spaceflight is that the onboard computational capabilities in spacecraft are still quite limited, due to the need for radiation-hardened onboard computers to survive the harsh deep space environment. As a result, deep-space spacecraft must still be cautiously controlled remotely from the ground by teams of engineers, using dedicated deep-space communications infrastructure of limited capacity. Thus, despite the falling cost-to-orbit and the advancement of space technology, operation costs stemming from such activities remain stubbornly high. To propel a scalable future for space exploration, there is a strong need for autonomous capability that can operate with limited computational resources and at a level of safety and reliability worthy for use on invaluable spacecraft.
The main goal of FAAST is to develop robust and computationally efficient algorithms for autonomous orbit guidance and control for low-cost space vehicles which must plan and re-plan their trajectories in an uncertain environment. This is achieved by the following four objectives:
1. Develop and validate a new approach that facilitates computationally feasible onboard techniques.
2. Ensure that the new formulation can be applied to the highly uncertain space environment.
3. Apply the new implementation to develop an algorithm for autonomous orbit guidance in a relevant test problem. A successful algorithm should be computationally lean and robust, facilitating on-board planning and re-planning.
4. High-fidelity testing of the algorithm and study of high-fidelity modeling of the space environment.
The first year of the project was dedicated largely to the development of a suitable new methodology for enabling computationally lean autonomous general problem-solving and decision making via onboard-ready optimization techniques. To this end, a new “monomial method” was devised, which rewrites nonlinear optimization problems in a simple way, but constraining the problem to a manifold, trading all the original problem nonlinearities for a tractable manifold constraint. Much of the first year also explored pre-computing state-related nonlinear relationships of interest using advanced techniques (namely state transition tensors; STTs, and differential algebra). This was applied to the problem of spacecraft rendezvous in low-Earth orbit and cislunar space.
The second year focused mainly on generalization of the monomial method, its applications in a range of onboard optimization problems of relevance in astrodynamics and spaceflight, and the improvement of the method leveraging techniques for reducing the complexity of the representation of a given optimization problem without compromising solution accuracy. Also performed in the end of the first and much of the second year were high-fidelity studies of the dynamics at work in binary asteroid systems, which are popular targets for deep space spacecraft. In this capacity the project interfaced with the ERC TRACES project and made use of software developed in the GRAINS MSCA project (Grant IDs 101077758 and 800060, respectively).
Overall the work resulted in 3 journal articles (1 published, 2 submitted), 5 conference works, and 3 software prototypes (one related to dynamical modeling and two related to spacecraft guidance).
The second year focused mainly on generalization of the monomial method, its applications in a range of onboard optimization problems of relevance in astrodynamics and spaceflight, and the improvement of the method leveraging techniques for reducing the complexity of the representation of a given optimization problem without compromising solution accuracy. Also performed in the end of the first and much of the second year were high-fidelity studies of the dynamics at work in binary asteroid systems, which are popular targets for deep space spacecraft. In this capacity the project interfaced with the ERC TRACES project and made use of software developed in the GRAINS MSCA project (Grant IDs 101077758 and 800060, respectively).
Overall the work resulted in 3 journal articles (1 published, 2 submitted), 5 conference works, and 3 software prototypes (one related to dynamical modeling and two related to spacecraft guidance).
The general methodology of pre-computing complex problem nonlinearities of interest for optimization problems represents a significant step forward in recently popular methods of trading memory for processor speed for on-board problem solving. However, existing methods generally introduce nonlinear corrections to traditionally linear onboard algorithms in an ad-hoc fashion. This work by contrast provides a general nonlinear framework with promising paths forward for both optimal guidance in uncertain environments and further investigation into the minimal fundamental functions which govern the guidance and control problems encountered in spaceflight. Provided with such information and a hierarchy for its manipulation, we can algorithmically generate lean representations of complex problems for onboard decision-making, essentially “compressing” complex optimization problems into computationally feasible versions that retain flexibility and reliability. The end goal of this is trustworthy autonomy for computationally constrained deep-space applications.