Option Discovery

Option discovery in reinforcement learning aims to automatically identify temporally extended actions (options) that improve learning efficiency and planning in complex environments. Current research focuses on developing algorithms that discover options maximizing both diversity and coverage of the state space, often employing techniques like Determinantal Point Processes, Kronecker graph approximations, and deep learning models such as variational autoencoders and deep Q-networks. These advancements are significant because they enable more efficient exploration and planning in multi-agent systems and continuous control tasks with sparse rewards, leading to improved performance in challenging reinforcement learning problems.