A High-efficient Training Strategy for Deep Q-learning Network Used in Robot Active Object Detection

Abstract

Active object detection (AOD) aims to guide a robot to make appropriate moving actions to get close to the target object, which is significant for the service robot to complete tasks in the indoor household environment. At present, most of the research on the robot AOD is developed based on reinforcement learning (RL) methods. However, the training efficiency and testing performance can be further improved. Therefore, a novel high-efficient training strategy is designed for the DQN model of AOD in this paper. Different from the existing RL-based training algorithm, the presented training strategy can avoid the repeat data with negative reward. The experiments have been implemented in an AOD dataset, proving that the proposed training strategy is more efficient than the raw DQN training algorithm.