Learning-based reconnaissance mission planning for communication-aware unmanned aerial vehicle system

Abstract

This thesis formulates a multi-target reconnaissance mission planning problem for UAV system with air-to-ground and air-to-air communication models. Then, a machine learning-based planning framework is proposed to obtain a near-optimal planning solution with fast runtime. The problem focuses on mission level planning, such as path planning and task planning under communication requirements during the mission period. Conventional optimization approaches for UAV mission planning problems have suffered from their high computational complexity, one of the main obstacles in developing near-optimal online planning algorithms. As an alternative, adequately trained deep neural networks for control/planning problems have been reported as fast and near-optimal online controllers/planners even with low onboard computation resources. This thesis aims to develop a learning-based framework for the communication-aware UAV mission planning problem. This thesis covers both single-UAV system with air-to-ground communication model and multi-UAV system with air-to-air/ground communication model. For a single-UAV system, this thesis proposes Markov decision process (MDP) formulation for a problem to plan the flight path and task sequence for reconnaissance of multiple targets under risk of communication loss to a ground base. A framework for training dataset generation, offline learning, and online planning is proposed. Various air-to-air communication models are considered for the multi-UAV system, including multi-hop communication, network connectivity constraint, and dynamic topology change. Then multi-UAV network mission planning problem is suggested to include issues on multi-UAV task assignment and multi-UAV communication network design problem. Both multi-agent reinforcement learning approach and optimization-based approach are proposed. Case studies on multi-target reconnaissance using single/multi-UAV have been conducted to demonstrate the effectiveness of the proposed approaches for mission environments under communication restrictions.