Artificial Intelligence techniques for GNC design, implementation and verification
Programme
TDE
Programme Reference
T705-605SA
Prime Contractor
CITY, UNIVERSITY of LONDON
Start Date
End Date
Status
Contracted
Country
United Kingdom
Objectives
To augment GNC systems by using Artificial Intelligence techniques in order to improve the performances, flexibility, autonomy and capability to handle failures and performance degradation of GNC systems while satisfying the reliability standards of a safety critical space system. To use AI techniques to design and analyse GNC systems.
Description
The performance of currently available Guidance Navigation and Control (GNC) systems for space is limited by unmodelled effects or uncertain parameters (e.g. sensor/actuator performances and performance degradations). Artificial Intelligence (AI) techniques based on machine learning could be integrated in GNC systems in order to enable on-line learning of the properties of dynamics, environment, sensors and actuators that limit the available performances. Such AI-assisted GNC system would increase both the performance and the degree of autonomy available on-board in terms of robustness, adaptability and awareness. However, typical AI techniques do not meet the reliability standards for safety critical space systems (e.g. reproducibility, robustness, computational efficiency, guaranteed convergence properties).
It is needed to investigate the criticalities and challenges of the use of AI techniques for GNC design, implementation and verification. In particular, the integration of AI techniques into a typical GNC control framework (covering classical as well as modern control techniques) shall be analysed.
The focus of the activity is on integrating techniques based on model knowledge (physical system as well as architectural configuration) with data-based on-line learning. Robustness and compatibility with or need for adaptation of the available control design and analysis frameworks and established qualification processes is to be assessed. The resulting AI-supported GNC system will be more flexible, adaptable and more performant than a classical GNC system, while still meeting the standards of a safety critical space system.
This activity entails the following tasks:
- review formal mathematical approaches to develop a robust and explainable AI technology
- establish the functional and performance requirements applicable to an AI-assisted GNC design process and to an AI-augmented GNC system
- perform trade-off of suitable mathematical AI approaches compatible with the current GNC architectures and design processes (model-based approach), including complexity, effort and expected benefits assessment
- establish the AI techniques suitable to the modelling, control and verification needs in the view of a robust and explainable AI-supported GNC architectures and functions
- develop a prototype set of benchmark problems for AI-assisted GNC design and AI-augment GNC system as well as for AI-supported autonomy (using an in-orbit assembly scenario including handling of failures and degradations)
- perform a detailed design and coding of the established AI techniques applied to AI-assisted GNC design and to AI-augment GNC system
- assess the performance and robustness of the AI-assisted GNC system
- define way forward for AI-GNC system deployment
Software shall be delivered under an ESA Software Community Licence, so that any individuals or entities within ESA Member States can access to it and can provide update to the community of users.
Application Domain
Generic Technologies
Technology Domain
5 - Space System Control
8 - System Design & Verification
Competence Domain
3-Avionic Systems
Initial TRL
TRL 1
Target TRL
TRL 3
Public Document
Executive Summary