Recovery from Adversarial Attacks in Cyber-physical Systems: Shallow, Deep and Exploratory Works

TL;DR

This paper defines recovery for cyber-physical systems (CPS) as online restoration of target physical states under adversarial attacks, formalized around a predicate $P$ over the state $s_t$. It surveys 30 papers (2014–2023) and organizes them into shallow (no dedicated recovery controller) and deep (with dedicated controllers) categories, plus exploratory work, to map techniques across applications, attack surfaces, and dynamics. The authors show that most work centers on detection or signal restoration rather than full recovery, with abundant methods targeting sensor and communication-channel attacks, and a smaller set addressing deep, controller-led recovery; they identify gaps in attack diagnosis and rigorous soundness guarantees. They propose six future directions, including scenario-specific recovery, measurable soundness, real-time latency reduction, resource efficiency, attack-diagnosis enhancements, and broader exploratory studies, arguing these are essential to deploying resilient CPS in critical domains. Overall, the paper provides a structured framework and a comprehensive inventory to guide researchers toward more robust and scalable CPS recovery strategies.

Abstract

Cyber-physical systems (CPS) have experienced rapid growth in recent decades. However, like any other computer-based systems, malicious attacks evolve mutually, driving CPS to undesirable physical states and potentially causing catastrophes. Although the current state-of-the-art is well aware of this issue, the majority of researchers have not focused on CPS recovery, the procedure we defined as restoring a CPS's physical state back to a target condition under adversarial attacks. To call for attention on CPS recovery and identify existing efforts, we have surveyed a total of 30 relevant papers. We identify a major partition of the proposed recovery strategies: shallow recovery vs. deep recovery, where the former does not use a dedicated recovery controller while the latter does. Additionally, we surveyed exploratory research on topics that facilitate recovery. From these publications, we discuss the current state-of-the-art of CPS recovery, with respect to applications, attack type, attack surfaces and system dynamics. Then, we identify untouched sub-domains in this field and suggest possible future directions for researchers.

Figures (10)

• Figure 1: Shallow vs. deep recovery methods.
• Figure 2: Vulnerabilities of CPS against adversarial attacks.
• Figure 3: Tree diagram of the surveyed literature.
• Figure 4: Workflow of component exclusion method discussed in shin2019rnnDetector. The recovery module assigns weights to redundant sensors based on how much each of them can be trusted. The assumption is multiple sensors are serving for the same function, i.e., redundancy.
• Figure 5: A typical state estimation-based shallow recovery. Upon a sensor detection, the system switches from using physical sensor readings to estimated states. A checkpointer kong2018checkpointing can be optionally used to accelerate state estimator by caching information.

Theorems & Definitions (3)

• definition 1: Target physical states
• definition 2: Recovery
• definition 3: Shallow and Deep Recovery