文献类型：期刊文献

英文题名：A modal separation deep reinforcement learning method for navigation of autonomous mobile robot

作者：Zhang, Xinke[1]; Wang, Liansheng[1]; Yi, Jianjun[1]

机构：[1] East China University of Science and Technology, School of Mechanical and Power Engineering, Shanghai, China

年份：2026

起止页码：1091

外文期刊名：2026 11th International Conference on Intelligent Computing and Signal Processing, ICSP 2026

收录：EI(收录号：20262520961752)

语种：英文

外文关键词：Deep learning - Deep reinforcement learning - Intelligent robots - Knowledge management - Knowledge transfer - Mobile robots - Navigation - Robot learning - Transfer learning

摘要：Existing robot navigation methods based on reinforcement learning suffer from slow convergence in complex environments. Also, the complexity of sim-to-real transfer of reinforcement learning is increased due to multimodal inputs. To address these issues, we propose a reinforcement learning architecture of modal separation learning (MSDRL) for mobile robot navigation. The training of modal separation divides navigation task into three stages and uses knowledge transfer to accelerate agent learning. A comparison study between our method, a deep learning-based method, and other deep reinforcement learning approaches demonstrate that our MSDRL performed better in simulation and real-world in terms of success rate, cumulative distance, and trajectory smoothness. Test results on real robot also show the navigation efficiency of MSDRL. ? 2026 IEEE.