Appearing Attacks on LiDAR-Based 3D Object Detectors

LiDAR is pivotal for the perception and safety of autonomous vehicles (AVs). Recent research indicates that the outputs of object detectors can be deliberately manipulated through carefully designed perturbations to the input data. This study introduces a novel attack methodology, termed the “appearing attack”, which induces object detectors to generate erroneous results by strategically incorporating specific features into the input data. To validate the feasibility and transferability of this attack, we conducted experiments using random parameter settings. Subsequently, we refined the approach through optimal parameter appearing attacks to maximize the effectiveness of this strategy.