Efficient autonomous exploration with dueling DDQN enhancing active SLAM through reinforcement learning
Publication Type
Conference Proceeding Article
Publication Date
11-2024
Abstract
Active Simultaneous Localization and Mapping (SLAM) is a critical problem in autonomous robotics, enabling robots to navigate to new regions while building an accurate model of their surroundings. This paper presents a novel Active SLAM framework enhanced by Reinforcement Learning (RL) techniques to address challenges in autonomous navigation within unknown environments. The proposed system introduces two key innovations: (1) the integration of Dueling Double Deep Q-Networks (Dueling DDQN) to dynamically select frontiers for exploration based on real-time environmental feedback, and (2) a multi-component reward function that optimally balances exploration and exploitation by incorporating factors such as distance, information gain, and revisitation penalties. These innovations result in a significant performance gain, with the Dueling DDQN method demonstrating up to a 45 percentage points improvement in map completion rates and a 35 percentage points reduction in exploration time compared to traditional SLAM and standard RL approaches. Additionally, a novel evaluation method is introduced, using image-based comparison techniques to assess performance in environments without ground truth data. Experimental results in simulated environments validate the system’s ability to enhance both mapping accuracy and computational efficiency, demonstrating the potential of RL to advance autonomous robotic exploration and mapping. The implementation and source code for this work can be found at https://github.com/LiXin0123/ExplORB-SLAM-Dueling-DDQN.git, providing a comprehensive resource for further experimentation and validation.
Keywords
active SLAM, frontier exploration, ORBSLAM, reinforcement learning, visual-graph SLAM
Discipline
Artificial Intelligence and Robotics
Publication
2024 9th International Conference on Robotics and Automation Engineering (ICRAE): Singapore, November 15-17: Proceedings
First Page
123
Last Page
130
ISBN
9798331518301
Identifier
10.1109/ICRAE64368.2024.10851611
Publisher
IEEE
City or Country
Pistacataway
Citation
LI, Xin; LEONG, Kenji Kah Hoe; SEOW, Chee Kiat; PUGEAULT, Nicholas; and CAO, Qi.
Efficient autonomous exploration with dueling DDQN enhancing active SLAM through reinforcement learning. (2024). 2024 9th International Conference on Robotics and Automation Engineering (ICRAE): Singapore, November 15-17: Proceedings. 123-130.
Available at: https://ink.library.smu.edu.sg/sis_research/10124
Additional URL
https://doi.org/10.1109/ICRAE64368.2024.10851611
