Small-Data-Driven Multitask Computational Guidance for Constrained Impact

Abstract

In traditional data-driven deep learning guidance, variations in guidance tasks may result in sharp performance degradation in terms of accuracy and efficiency of pretrained models. However, obtaining a guidance model adaptable to new tasks requires a substantial investment of training data and time. To address this issue, this paper proposes a small-data-driven multitask computational guidance (SMCG) algorithm utilizing a novel gated progressive neural network structure. The SMCG closely emulates the human learning process. Specifically, it progressively acquires knowledge from simple to complex tasks (from easy to difficult), applies previously learned knowledge from old tasks to new ones (reusing old knowledge), and achieves continuous learning for new tasks by adding new columns (sustainable learning). Simulation results demonstrate that SMCG can effectively reuse expertise acquired from previous tasks in new tasks, thereby reducing the required training data and enhancing the adaptability of the guidance model to new tasks.