Session
Technical Poster Session 3
Location
Utah State University, Logan, UT
Abstract
The focus of our initial application scenario centers around a low-thrust orbit transfer in Low-Earth Orbit (LEO). This specific use-case has been chosen due to its inherent challenges, including the requirements for robustness and real-time computation.
We propose an AI-based solution capable of autonomous and robust on-board G&C. The core of our approach leverages a Deep Neural Network (DNN) trained through Reinforcement Learning (RL) techniques. Our method aims at enhancing a traditional guidance approach by managing environmental perturbations, it processes the on-board navigation coordinates and provides the thrust to be imposed by the propulsion subsystem.
Our approach demonstrates effectiveness in performing maneuvers changing semi-major axis (SMA), eccentricity (ECC), and inclination (INC), operating continuously with a control horizon of several days. Robustness is tested by using physical model uncertainties, introducing disturbances in the mission coordinates, and injecting perturbations in subsystems.
Document Type
Event
Toward Autonomous Guidance and Control: A Robust AI-Based Solution for Low-Thrust Orbit Transfers
Utah State University, Logan, UT
The focus of our initial application scenario centers around a low-thrust orbit transfer in Low-Earth Orbit (LEO). This specific use-case has been chosen due to its inherent challenges, including the requirements for robustness and real-time computation.
We propose an AI-based solution capable of autonomous and robust on-board G&C. The core of our approach leverages a Deep Neural Network (DNN) trained through Reinforcement Learning (RL) techniques. Our method aims at enhancing a traditional guidance approach by managing environmental perturbations, it processes the on-board navigation coordinates and provides the thrust to be imposed by the propulsion subsystem.
Our approach demonstrates effectiveness in performing maneuvers changing semi-major axis (SMA), eccentricity (ECC), and inclination (INC), operating continuously with a control horizon of several days. Robustness is tested by using physical model uncertainties, introducing disturbances in the mission coordinates, and injecting perturbations in subsystems.
