This paper studies how to improve the generalization performance and learning speed of the navigation agents trained with deep reinforcement learning (DRL). DRL exhibits huge potential in mapless navigation, but DRL agents performing well in training scenarios are found to perform poorly in unfamiliar real-world scenarios. In this work, we present the representation of LiDAR readings as a key factor behind agents performance degradation and propose a simple but powerful input pre-processing (IP) approach to improve the agents performance. As this approach uses adaptively parametric reciprocal functions to pre-process LiDAR readings, we refer to this approach as IPAPRec and its normalized version as IPAPRecN. IPAPRec/IPAPRecN can highlight important short-distance values and compress the range of less-important long-distance values in laser scans, which well addressed the issues induced by conventional representations of laser scans. Their high performance is validated by extensive simulation and real-world experiments. The results show that our methods can substantially improve agents success rates and greatly reduce the training time compared to conventional methods.
تحميل البحث