Abstract
Quantum computing (QC) has received a lot of attention according to its light training parameter numbers and computational speeds by qubits. Moreover, various researchers have tried to enable quantum machine learning (QML) using QC, where there are also multifarious efforts to use QC to implement quantum multi-agent reinforcement learning (QMARL). Existing classical multi-agent reinforcement learning (MARL) using neural network features non-stationarity and uncertain properties due to its large number of parameters. Therefore, this paper presents a visual simulation software framework for a novel QMARL algorithm to control autonomous multi-drone systems to take advantage of QC. Our proposed QMARL framework accomplishes reasonable reward convergence and service quality performance with fewer trainable parameters than the classical MARL. Furthermore, QMARL shows more stable training results than existing MARL algorithms. Lastly, our proposed visual simulation software allows us to analyze the agents' training process and results.
Abstract (translated)
URL
https://arxiv.org/abs/2211.15375
