Research on Autonomous Driving Element Technology based on Low Performance Embedded PC for Micro Lunar Exploration Rover

In this study, we selected lightweight AI models such as LeNet, miniVGG-Net, Shallow-Net, AlexNet, MobileNet, and GoogLeNet. These models have been recently applied or considered for CubeSat space missions. The goal of this study is to identify SOTA (State-Of-The-Art) models that could be considered for use in implementing autonomous driving in extraterrestrial environments such as the Moon or Mars. The classification performance of these models was analyzed in a categorical classification problem, including label classes such as moving straight, turning right, and turning left. The results showed that the AlexNet model had the highest performance, with an ACC (Accuracy) of 0.9999 and an F-1 score of 0.9999, while the MobileNet model had the lowest performance, with an ACC of 0.8000 and an F-1 score of 0.4572. Consequently, AlexNet and LeNet were selected as the benchmarks for comparing and analyzing the performance of RoverNet-1, the AI for autonomous driving to be developed for future exploration rovers.