Purpose

This paper aims to use the Monodepth method to improve the prediction speed of identifying the obstacles and proposes a Probability Dueling DQN algorithm to optimize the path of the agent, which can reach the destination more quickly than the Dueling DQN algorithm. Then the path planning algorithm based on Probability Dueling DQN is combined with FastSLAM to accomplish the autonomous navigation and map the environment.

Design/methodology/approach

This paper proposes an active simultaneous localization and mapping (SLAM) framework for autonomous navigation under an indoor environment with static and dynamic obstacles. It integrates a path planning algorithm with visual SLAM to decrease navigation uncertainty and build an environment map.

Findings

The result shows that the proposed method offers good performance over existing Dueling DQN for navigation uncertainty under the indoor environment with different numbers and shapes of the static and dynamic obstacles in the real world field.

Originality/value

This paper proposes a novel active SLAM framework composed of Probability Dueling DQN that is the improved path planning algorithm based on Dueling DQN and FastSLAM. This framework is used with the Monodepth depth image prediction method with faster prediction speed to realize autonomous navigation in the indoor environment with different numbers and shapes of the static and dynamic obstacles.

中文翻译：

---

用于室内环境自主导航的概率对决 DQN 主动视觉 SLAM

目的

本文旨在利用Monodepth方法来提高识别障碍物的预测速度，并提出一种Probability Dueling DQN算法来优化agent的路径，比Dueling DQN算法可以更快地到达目的地。然后将基于概率决斗DQN的路径规划算法与FastSLAM相结合，完成自主导航和环境地图绘制。

设计/方法/方法

本文提出了一种主动同时定位和映射（SLAM）框架，用于在具有静态和动态障碍物的室内环境下进行自主导航。它将路径规划算法与视觉 SLAM 相结合，以减少导航不确定性并构建环境地图。

发现

结果表明，所提出的方法在现实世界中静态和动态障碍物的数量和形状不同的室内环境下，对于导航不确定性，所提出的方法比现有的Dueling DQN具有良好的性能。

原创性/价值

本文提出了一种新的主​​动SLAM框架，它由概率决斗DQN组成，是基于决斗DQN和FastSLAM的改进路径规划算法。该框架与具有更快预测速度的Monodepth深度图像预测方法配合使用，实现在静态和动态障碍物数量和形状不同的室内环境中的自主导航。