Scholarlink

<p>This paper studies the mapless point navigation problem (MPNP) in the indoor environment. However, conventional navigation methods can not solve the MPNP efficiently due to the lack of prior information. We propose a goal-driven navigation method based on the deep reinforcement learning approach with a newly designed reward mechanism. In addition, a deep network ResNet18 is introduced to extract scene information. At the same time, the proximal policy optimization (PPO) algorithm is employed to assist the robot in obstacle avoidance. The experiment utilizes Unreal Engine 5 to create the experimental scene and analyzes the feasibility of the reinforcement learning algorithm. The training results demonstrate that this method yields excellent experimental performance and good generalization.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Efficient Mapless Point Navigation in Indoor Environments Using Goal-Driven Deep Reinforcement Learning

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Tai Zhang,
  • Chenglei Xia,
  • Jijiang Xu,
  • Liping Bai

摘要

This paper studies the mapless point navigation problem (MPNP) in the indoor environment. However, conventional navigation methods can not solve the MPNP efficiently due to the lack of prior information. We propose a goal-driven navigation method based on the deep reinforcement learning approach with a newly designed reward mechanism. In addition, a deep network ResNet18 is introduced to extract scene information. At the same time, the proximal policy optimization (PPO) algorithm is employed to assist the robot in obstacle avoidance. The experiment utilizes Unreal Engine 5 to create the experimental scene and analyzes the feasibility of the reinforcement learning algorithm. The training results demonstrate that this method yields excellent experimental performance and good generalization.