GAN-Based Robust Motion Planning for Mobile Robots Against Localization Attacks
Wenbing Tang,Yuan Zhou,Haiying Sun,Yuhong Zhang,Yang Liu,Zuohua Ding,Jing Liu,Jifeng He
DOI: https://doi.org/10.1109/lra.2023.3241807
IF: 5.2
2023-03-01
IEEE Robotics and Automation Letters
Abstract:Motion planning (MP) is essential but challenging for mobile robots. Most of the existing MP methods, at each instant, compute an action based on the states of the robot and the surrounding obstacles, assuming that the robot's localization module is attack-free. Unfortunately, the localization module is vulnerable to sensor attacks, such as GPS spoofing attacks. In this letter, we propose a novel robust framework, GAN-MP, where a generative adversarial network (GAN) is exploited to mitigate the localization attacks, and the state-of-the-art MP methods are applied to generate collision-free actions. Specifically, GAN-MP aims to learn a Generator to compute the potential positions of the robot. Consequently, it can reserve the robot's benign states while correcting the attacked states. Hence, it is suitable for benign and attacked scenarios without any attack detector. In addition, GAN-MP is method-agnostic and can be easily integrated with any existing MP method. We instantiate GAN-MP with a deep reinforcement learning method to demonstrate its design and training processes. Comprehensive experiments show that GAN-MP can mitigate localization attacks and guarantee safe motion. We also demonstrate the robustness and generalization of GAN-MP.
robotics