Deep reinforcement learning spacecraft guidance with state uncertainty for autonomous shape reconstruction of uncooperative target

• DOI: 10.1016/j.asr.2023.07.007
• Publication Date: 2023-07-07T01:29:02Z
• Authors: Andrea Brandonisio; Lorenzo Capra; Michèle Lavagna

In recent years, space research has shifted heavily its focus towards enhanced autonomy on-board spacecrafts for on-orbit servicing activities (OOS). OOS and proximity operations include a variety of activities: the focal point this work is autonomous guidance chaser spacecraft shape reconstruction an artificial uncooperative object. Adaptive depends on ability system to build map uncertain environment, figuring out location inside it accordingly determining control law. Thus, navigation framed as active Simultaneous Localization Mapping (SLAM) problem modeled Partially Observable Markov Decision Process (POMDP). A state-of-the-art Deep Reinforcement Learning (DRL) method, Proximal Policy Optimization (PPO), investigated develop agent capable cleverly planning target. Starting from previous topic, develops further proposing continuous action space, such that no more forced choose between predefined set possible discrete actions, fixed both in magnitude direction. way any combination three-dimensional thrust vector components available. The small satellite mounting electric propulsion engine defining range, linearized eccentric relative motion with selected Through rendered triangular mesh, capabilities geometry mapping are evaluated, considering number quality pictures made each face. Extensive training tests performed random initial conditions verify generalizing capability DRL agent. results then validated comprehensive testing campaign, whose primary introduction noisy measurements coming navigation, affecting pose estimation. sensitivity proposed method condition analyzed efficiency retraining procedure examined. applicability methods neural networks support close scenario corroborated technique employed vastly tested verified.