A Satellite Task Planning Algorithm Based on a Symmetric Recurrent Neural Network

Liu, Sikai; Yang, Jun

doi:10.3390/sym11111373

Cited by 9 publications

(3 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Machine learning has shown some advantages in satellite mission planning and scheduling problems [ 14 , 15 , 16 ]. Deep reinforcement learning is an effective method for solving the sequential decision problem.…”

Section: Introductionmentioning

confidence: 99%

An LEO Constellation Early Warning System Decision-Making Method Based on Hierarchical Reinforcement Learning

Cheng

Wei

Sun

et al. 2023

Sensors

View full text Add to dashboard Cite

The cooperative positioning problem of hypersonic vehicles regarding LEO constellations is the focus of this research study on space-based early warning systems. A hypersonic vehicle is highly maneuverable, and its trajectory is uncertain. New challenges are posed for the cooperative positioning capability of the constellation. In recent years, breakthroughs in artificial intelligence technology have provided new avenues for collaborative multi-satellite intelligent autonomous decision-making technology. This paper addresses the problem of multi-satellite cooperative geometric positioning for hypersonic glide vehicles (HGVs) by the LEO-constellation-tracking system. To exploit the inherent advantages of hierarchical reinforcement learning in intelligent decision making while satisfying the constraints of cooperative observations, an autonomous intelligent decision-making algorithm for satellites that incorporates a hierarchical proximal policy optimization with random hill climbing (MAPPO-RHC) is designed. On the one hand, hierarchical decision making is used to reduce the solution space; on the other hand, it is used to maximize the global reward and to uniformly distribute satellite resources. The single-satellite local search method improves the capability of the decision-making algorithm to search the solution space based on the decision-making results of the hierarchical proximal policy-optimization algorithm, combining both random hill climbing and heuristic methods. Finally, the MAPPO-RHC algorithm’s coverage and positioning accuracy performance is simulated and analyzed in two different scenarios and compared with four intelligent satellite decision-making algorithms that have been studied in recent years. From the simulation results, the decision-making results of the MAPPO-RHC algorithm can obtain more balanced resource allocations and higher geometric positioning accuracy. Thus, it is concluded that the MAPPO-RHC algorithm provides a feasible solution for the real-time decision-making problem of the LEO constellation early warning system.

show abstract

Section: Introductionmentioning

confidence: 99%

An LEO Constellation Early Warning System Decision-Making Method Based on Hierarchical Reinforcement Learning

Cheng

Wei

Sun

et al. 2023

Sensors

View full text Add to dashboard Cite

show abstract

“…Mission planning aims to allocate several tasks considering high-level goals, constraints, and resources. In recent years, researchers started studying this topic from different angles: using fuzzy neural networks to respond to uncertainty and proposing rescheduling [19], exploiting symmetric neural networks to improve heuristic search [20], using tabular search algorithms [21]. All these methods fall into the category of Static Scheduling because of the time they require to gather all the tasks.…”

Section: Introductionmentioning

confidence: 99%

An Application of Online Learning to Spacecraft Memory Dump Optimization

Cesari¹,

Jonathan²,

Maestrini³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Song et al designed an autonomous ground planner to acquire ground targets according to emergency levels [15]. An even more cutting-edge solution is proposed in [16], where S. Liu and J. Yang designed a satellite task planner by means of a heuristic search algorithm based on a symmetric recurrent neural network. Many other works can be listed as interesting solutions to the problem, such as [17][18][19][20][21][22][23][24][25].…”

Section: Introductionmentioning

confidence: 99%

Analytical Attitude Guidance Planner for Multiple Ground Targets Acquisitions

2022

View full text Add to dashboard Cite

This paper focuses on the development of a guidance methodology for the planning of multiple ground target acquisition. Specifically, the work addresses the problem of the lack of an attitude guidance planner (AGP) aboard a remote sensing satellite. In general, the guidance is computed offline and uploaded by ground control to the space segment, i.e., satellites are not responsible for the guidance generation but they only perform control algorithms to track the guidance profiles provided by the ground segment. Overall, this limits the mission flexibility and efficiency, affecting the capability of autonomous satellite decisions. This choice is driven by the fact that the numerical algorithms used to optimize the attitude guidance trajectory require high computational effort to be implemented directly on the satellite computer. Therefore, the aim of this work is to design an analytical AGP solution to solve this problem by requiring low computational effort, making it suitable for real-time applications on on-board flight hardware. In this way, the satellite’s guidance, navigation, and control (GNC) module would become completely autonomous and independent of ground control, which will only have to indicate the targets to be acquired so that the satellite can generate its own guidance for the GNC module. The AGP analytical solution for multiple ground target acquisition is evaluated by means of phases: the first phase is named the APPG (attitude point-to-point generator) and it aims to generate the point-to-point guidance to start the ground target acquisition. The second phase is named the ATPG (attitude target pointing generator) and it generates the reference guidance to maintain the payload view axis pointing toward the ground target. The two phases joined together give the whole guidance needed to observe ground target points by means of an analytical closed-form solution.

show abstract

A Satellite Task Planning Algorithm Based on a Symmetric Recurrent Neural Network

Cited by 9 publications

References 16 publications

An LEO Constellation Early Warning System Decision-Making Method Based on Hierarchical Reinforcement Learning

An LEO Constellation Early Warning System Decision-Making Method Based on Hierarchical Reinforcement Learning

An Application of Online Learning to Spacecraft Memory Dump Optimization

Analytical Attitude Guidance Planner for Multiple Ground Targets Acquisitions

Contact Info

Product

Resources

About