Adaptable Automation with Modular Deep Reinforcement Learning and Policy Transfer

Recent advances in deep Reinforcement Learning (RL) have created\nunprecedented opportunities for intelligent automation, where a machine can\nautonomously learn an optimal policy for performing a given task. However,\ncurrent deep RL algorithms predominantly specialize in a narrow range of tasks,\nare sample inefficient, and lack sufficient stability, which in turn hinder\ntheir industrial adoption. This article tackles this limitation by developing\nand testing a Hyper-Actor Soft Actor-Critic (HASAC) RL framework based on the\nnotions of task modularization and transfer learning. The goal of the proposed\nHASAC is to enhance the adaptability of an agent to new tasks by transferring\nthe learned policies of former tasks to the new task via a "hyper-actor". The\nHASAC framework is tested on a new virtual robotic manipulation benchmark,\nMeta-World. Numerical experiments show superior performance by HASAC over\nstate-of-the-art deep RL algorithms in terms of reward value, success rate, and\ntask completion time.\n