8964 - APPLICATION OF LSTM NETWORKS TO REAL-TIME PREDICTION OF A WHEEL TRAVELING PHENOMENA

Paper presented at ISTVS 2025 | 55th Conference of the International Society for Terrain-Vehicle Systems https://doi.org/10.56884/DUMF7XJZ

Authors: Haruki Endo, Shingo Ozaki, Kai Xing

Keywords: Wheel; Trafficability; Simulation; Deep learning; LSTM network

Abstract:

Lunar exploration, which began in the 1960s, is still ongoing and various space exploration programs are planned. In these space exploration programs, it is important to analyze and predict the trafficability of exploration rovers on the lunar and Martian surfaces, and to develop and operate on these ground conditions. Therefore, in this study, we focused on the real-time analysis in operational phase of exploration rovers. We applied a Long Short-term Memory (LSTM)1) network, an artificial regression neural network architecture used in the field of deep learning, to single-wheel traveling phenomena with the aim of applying it to the real-time prediction in future exploration rover operations. Specifically, numerical simulations were carried out using an extended terramechanics model that takes terrain surface deformation into account, and training and prediction are carried out on several LSTM models. The results show the possibility of real-time prediction of trafficability of a wheel on soft ground due to the characteristics of LSTM networks that can handle time-series data. Furthermore, it is suggested that the proposed method may be applicable to the prediction of precursor phenomena of wheel stuck.

1. Gonzalez, J ; Yu, W. Non-linear system modeling using LSTM neural networks. IFAC-Pap. 2018, 51, 485-489.