Safety Enhancement in Planetary Rovers: Early Detection of Tip-over Risks Using Autoencoders

摘要

Autonomous robots consistently encounter unforeseen dangerous situations during exploration missions. The characteristic rimless wheels in the AsguardIV rover allow it to overcome challenging terrains. However, steep slopes or difficult maneuvers can cause the rover to tip over and threaten the completion of a mission. This work focuses on identifying early signs or initial stages for potential tip-over events to predict and detect these critical moments before they fully occur, possibly preventing accidents and enhancing the safety and stability of the rover during its exploration mission. Inertial Measurement Units (IMU) readings are used to develop compact, robust, and efficient Autoencoders that combine the power of sequence processing of Long Short-Term Memory Networks (LSTM). By leveraging LSTM-based Autoencoders, this work contributes predictive capabilities for detecting tip-over risks and developing safety measures for more reliable exploration missions.