Detection Is Not Enough - Attack Recovery for Safe and Robust Autonomous Robotic Vehicles

Explore a 15-minute conference talk from USENIX Enigma 2022 that delves into attack recovery strategies for autonomous robotic vehicles (RVs). Learn about the limitations of attack detection alone and discover two novel frameworks designed to provide safe responses to attacks, allowing RVs to continue their missions despite malicious interventions. Understand the importance of sensor measurements in RV perception and the vulnerabilities they present. Examine the Feed-Forward Controller (FFC) technique that monitors and takes over when attacks are detected, as well as a method for identifying and isolating compromised sensors while using historic states to estimate current RV positioning. Gain insights into the challenges of GPS spoofing, sensor tampering, and their potential impacts on military drones and marine navigation systems.

Intro
Perception in Robotic Vehicles (RV)
Sensor Attacks Against Robotic Vehicles (RV)
Attack Detection, Anomaly Detection
Failsafe is not enough either...
Sensor → PID Control → Actuator Signal
RV under Attack
PID Over-Compensates under Attacks
Approach to design Recovery Techniques
Feedforward Controller (FFC) Design
Recovery Framework
Multiple Sensor under Attack
Attack Setting
Recovery Goal
Prevent Erroneous Actuator Signals
Identify the Sensor(s) under attack
Isolate Sensor(s) from Control Process
Substitute Input Sequence: Historic States
Recovery with Historic State Input
Without Recovery