A novel robotic gas leak source localization method in lunar base using deep reinforcement learning algorithm
- Paper ID
102370
- DOI
- author
- company
School of aeronautics and astronautics, Sun Yat-Sen University Guangzhou
- country
China
- year
2025
- abstract
The Moon, as the closest celestial body to Earth, served as the first step in humanity's exploration of the universe and extraterrestrial colonization. In the near future, humans will return to the Moon and establish a Moon base, which will play a crucial role in lunar exploration by providing life support for astronauts and protecting them from the harsh lunar environment. As a large-scale architectural structure, the Moon base, like buildings on Earth and the International Space Station (ISS), also faced the risk of toxic gas leaks, which posed a significant threat to astronauts' health and safety. The 1995 Tokyo subway sarin attack resulted in nearly 1,000 deaths, and the 2012 Gumi hydrogen fluoride gas leak caused multiple injuries. On the ISS, the Thermal Control System used ammonia as a coolant, and a system failure could have led to ammonia leakage into the cabin atmosphere. Similarly, this risk also existed in future lunar bases, making the rapid localization of leakage sources crucial in case of an accident. In lunar exploration missions, human resources were extremely valuable, and manually searching for gas leaks could have endangered astronauts. To address this issue, this paper proposed a robotic gas leakage source localization algorithm based on deep reinforcement learning. The searching process was formulated as a Markov Decision Process, and a Double Deep Q-Network algorithm was employed for leakage source localization. A 3D convolutional neural network was designed to estimate the values of the Q-function. The proposed algorithm was trained and evaluated using simulation data obtained from computational fluid dynamics software. These data were generated in the simulation of a simplified model of the bioregenerative life support system, Lunar Palace-1. Experimental results demonstrated that our algorithm effectively located the leakage source and outperformed two other widely used algorithms.