Participants: LTU Division of Space technology and OHB Sweden.
Objective (What and why)
In general, space missions operate in challenging environments with complex time varying dynamics that are difficult or incompletely modelled (for eg., during refuelling missions, the inertia tensor of both the spacecraft undergo significant changes due to changing fuel mass which may result in incorrect modelling leading to failures). The performance of the traditional control approaches based on optimization methods are majorly dependant on the accuracy of the underlying dynamics which limits their use in environments with partially known dynamics, gravitational parameters, unforeseen disturbances, or nonlinearities. Thus, there is a need for more robust and adaptive approaches which can address issues of changing dynamics, partially known environment or limited onboard memory and computational capability.
With the technological progress in machine learning, space organisations across the globe are looking forward to an emerging paradigm shift approach that will enable space autonomy empowered with intelligent, resilient architecture for the next-generation space mission. In case of mega-satellite constellation management, progressive optimization-based numerical optimal guidance algorithms in centralized as well as distributed framework have been explored. However, the scope of such optimization based solution approach has been extremely limited, in terms of capability of being a deployable solution for onboard implementation. On contrary, ML-based solutions are envisioned to be implemented onboard, since typically the computationally rigorous training processes are performed offline before deploying the solution for the actual implementation. Moreover, the generalization capabilities of machine learning approaches that can address such issues are gaining more attention.
The main goal of the OPTACOM project is to develop and verify algorithms enabling the functional autonomy of Low Earth Orbit (LEO) satellite constellation defining the avionics and attitude and orbit control system functions (e.g. guidance, cluster and orbit control, avoidance maneuvering, etc.) whilst being optimized with an on-board implementation of machine learning and embedded real-time optimization. The project focuses on developing a decentralized orbit control management algorithms for large satellite constellations to perform several autonomous tasks, such as orbit initialization (phasing) after launch and early orbit phase, station-keeping and collision avoidance. Our Robotics and AI team at LTU will contribute with the investigation to conceptualize and formulate the overall autonomy stack by exploiting machine learning based algorithms.
Sumeet Gajanan Satpute