Robust Adversarial Attacks Detection based on Explainable Deep Reinforcement Learning For UAV Guidance and Planning

Abstract

The danger of adversarial attacks to unprotected Uncrewed Aerial Vehicle (UAV) agents operating in public is growing.Adopting AI-based techniques and more specifically Deep Learning (DL) approaches to control and guide these UAVs can be beneficial in terms of performance but add more concerns regarding the safety of those techniques and their vulnerability against adversarial attacks causing the chances of collisions going up as the agent becomes confused. This paper proposes an innovative approach based on the explainability of DL methods to build an efficient detector that will protect these DL schemes and thus the UAVs adopting them from potential attacks. The agent is adopting a Deep Reinforcement Learning (DRL) scheme for guidance and planning. It is formed and trained with a Deep Deterministic Policy Gradient (DDPG) with Prioritised Experience Replay (PER) DRL scheme that utilises Artificial Potential Field (APF) to improve training times and obstacle avoidance performance. The adversarial attacks are generated by Fast Gradient Sign Method (FGSM) and Basic Iterative Method (BIM) algorithms and reduced obstacle course completion rates from 80% to 35%. A Realistic Synthetic environment for UAV explainable DRL based planning and guidance including obstacles and adversarial attacks is built. Two adversarial attack detectors are proposed. The first one adopts a Convolutional Neural Network (CNN) architecture and achieves an accuracy in detection of 80%. The second detector is developed based on a Long Short Term Memory (LSTM) network and achieves an accuracy of 91% with much faster computing times when compared to the CNN based detector.

Security Considerations in AI-Robotics: A Survey of Current Methods, Challenges, and Opportunities

“…Similarly, Bhandarkar et al [114] proposed DRL techniques to launch Sybil attacks and transmit spoofed beacon signals to disrupt the path planning logic. Hickling et al [115], on the other hand, proposed to utilize the explainability of DL methods. In their approach, the planning agent is trained with a Deep Deterministic Policy Gradient (DDPG) with Prioritized Experience Replay (PER) DRL scheme, that utilizes Artificial Potential Field (APF) to improve training time and obstacle avoidance performance.…”

Fragility Impact of RL Based Advanced Air Mobility under Gradient Attacks and Packet Drop Constraints

“…Threat Models Impact [11] Continuous disturbance signal with Gaussian noise Inability to estimate the flight path of the targeted UAV [18] FGSM and BIM Collision risk and reaching goals [19] Pixel level attack and semantic perturbation Object detection for UAV navigation [13] Triggerless backdoor attack on the model parameter UAV offloading policy [20] Adding imperceptible perturbations to the image Targeted UAV tracking [21] Adversarial attack based on forward derivative and optimization Navigation and control of UAV [22] Manipulate and control the input channel of the sensor. Collision with the obstacles.…”