V3I8P15

Robust Deep Reinforcement Learning-Based Guidance for Maneuvering UAVs under Observation Uncertainty

Le Trung Hoa¹*

Abstract

The increasing use of unmanned aerial vehicles (UAVs) with diverse and maneuverable flight behaviors presents new challenges for autonomous interception systems, particularly when the available observation information is affected by noise and uncertainty. Conventional model-based guidance methods generally rely on predefined motion models and accurate state information, which may limit their adaptability under uncertain operating conditions. This paper proposes a robust deep reinforcement learning (DRL)-based adaptive guidance framework for maneuvering UAV interception under observation uncertainty. The interception problem is formulated as a partially observable sequential decision-making process, in which a DRL agent learns an adaptive policy from relative motion observations. A multi-objective reward mechanism is designed to simultaneously consider task effectiveness, trajectory stability, and control smoothness. To improve robustness and generalization, uncertainty-aware training and domain randomization are incorporated into the learning environment, allowing the policy to experience a wide range of observation disturbances and target-motion conditions during training. The proposed approach is evaluated through statistically independent simulation scenarios and compared with conventional and standard DRL-based baselines. Performance is assessed in terms of task success, terminal error, robustness to observation uncertainty, generalization to previously unseen maneuvering conditions, and computational efficiency. The proposed framework provides a systematic approach for investigating robust learning-based adaptive guidance and offers a potential foundation for developing intelligent autonomous guidance methods in uncertain environments.

Keywords:

Deep reinforcement learning; adaptive guidance; UAV; observation uncertainty; robust learning; domain randomization; autonomous interception.