LSTM Noise Robustness: A Case Study for Heavy Vehicles

Artificial intelligence (AI) techniques are becoming more and more widespread. This is directly related to technology progress and aspects as the flexibility and adaptability of the algorithms considered, key characteristics that allow their use in the most variegated fields. Precisely the increasing diffusion of these techniques leads to the necessity of evaluating their robustness and reliability. This field is still quite unexplored, especially considering the automotive sector, where the algorithms need to be prepared to answer noise problems in data acquisition. For this reason, a methodology directly linked to previous works in the heavy vehicles field is presented. In particular, the same is focused on the estimation of rollover indexes, one of the main issues in road safety scenarios. The purpose is to expand the cited works, addressing the LSTM networks performance in case of strongly disturbed signals.