Model-Based DRL for Task Scheduling in Dynamic Environments for Cognitive Multifunction Radar

The uncertainty in the radar environment brings significant challenges to task scheduling in a cognitive multifunction radar (MFR). The recent radar task scheduling approaches assume the knowledge of the environment, which is unknown in real scenarios. To achieve online task scheduling for a cognitive radar which does not need to know the dynamics of the environment, this work investigates the model-based deep reinforcement learning (DRL), which learns the model of the environment to plan for scheduling tasks. The main idea of the model-based methods is to construct an abstract Markov-decision process (MDP) model such that planning in the abstract MDP is equivalent to planning in the real environment. Here, we tailor MuZero, proposed to learn to play games, to provide the needed model. The proposed algorithm is shown to be effective in MFR task scheduling while adapting to dynamic radar environments, without any a priori knowledge.