Path planning for autonomous mobile robot using transfer learning-based Q-learning

Transfer learning is the process of reusing the experience of agents in source tasks to improve the performance in new target tasks. In recent years, transfer learning has received more and more attention over the reinforcement learning settings. However, when applied to reinforcement learning, many problems will be exposed, such as how is the target task different from the source task, if the mappings between the tasks are required and what knowledge is transferred. Transfer learning algorithms have mainly been applied in discrete gridworld tasks. We first introduce the traditional Q-learning to the transfer algorithm based on the measurement of distance between two MDPs. Further, inspired by Q(lambad) algorithm, this paper investigates the improved Q-learning transfer algorithm to improve the learning efficiency. Finally, the simulation is shown to verify the effectiveness of the proposed algorithms.