In a modern cyber-physical system (CPS)—whether it is a power grid, an autonomous drone, or an industrial robot—the controller relies on a stream of data from sensors to understand the world. But what happens when some of those sensors are not just noisy, but maliciously compromised?
Traditional fault tolerance handles "random" failures, but a sophisticated adversary can inject false data that mimics legitimate system behavior, bypassing static detection filters. The research direction spearheaded by researchers like Fabio Pasqualetti, Yasser Shoukry, and Paulo Tabuada focuses on Secure State Estimation: the ability to reconstruct the true state of a system even when an unknown subset of its sensors and actuators has been hijacked.
The Combinatorial Wall
The fundamental difficulty of secure state estimation is not just detecting that something is wrong, but identifying which sensors are lying. If a system has $n$ sensors and an attacker can compromise up to $k$ of them, there are $\binom{n}{k}$ possible subsets of compromised sensors.
As Fabio Pasqualetti demonstrates in Attack Detection and Identification in Cyber-Physical Systems -- Part II: Centralized and Distributed Monitor Design, while detecting an anomaly is often straightforward, pinpointing the exact set of compromised components is an NP-hard problem. This "combinatorial explosion" means that as a power grid or sensor network grows, a brute-force search for the truth becomes computationally impossible.
Physics as the Source of Truth
To overcome this, researchers leverage the physics of the system as a consistency check. In a dynamic system, the state at time $t+1$ is constrained by the state at time $t$ and the laws of motion.
In Cyber-physical attacks in power networks: Models, fundamental limitations and monitor design, Fabio Pasqualetti uses geometric control theory to show that the physical interconnections of a power grid create a unique "signature" for legitimate data. If an attacker injects data that doesn't align with these dynamics, the physics of the network effectively "snitches" on the compromised nodes.
For more complex, non-linear systems like drones, Alberto L. Sangiovanni-Vincentelli and Yasser Shoukry apply this logic to "differentially flat" systems in A Satisfiability Modulo Theory Approach to Secure State Reconstruction in Differentially Flat Systems Under Sensor Attacks. By mapping complex non-linear trajectories into a simpler space, they can identify which sensor streams are physically impossible given the system's known capabilities.
Navigating the Search Space with SMT
Since we cannot check every combination of sensors, the field has moved toward borrowing tools from computer science: Satisfiability Modulo Theory (SMT) solvers.
In Secure State Estimation Under Sensor Attacks: A Satisfiability Modulo Theory Approach, Sanjit A. Seshia and his colleagues frame the estimation problem as a massive logic puzzle. The SMT solver doesn't just guess; it uses formal logic to prune large branches of the search space, quickly discarding combinations of sensors that are mathematically inconsistent with the observed physics.
This approach is further refined by Miroslav Pajic in A Scalable and Optimal Graph-Search Method for Secure State Estimation, which uses graph-based search to find the "largest set of consistent sensors," effectively isolating the attackers in real-time.
From Detection to Resilient Control
The ultimate goal is not just to see the attack, but to survive it. Recent work by Andrew Clark and Luyao Niu in LQG Reference Tracking With Safety and Reachability Guarantees Under Unknown False Data Injection Attacks moves toward "closed-loop" security. By maintaining a bank of potential estimators and solving optimization problems (QCQPs) on the fly, they ensure that even if some sensors are lying, the system's controller can still guarantee the vehicle stays within safe bounds.
Similarly, in the context of infrastructure, Ram Rajagopal and Yue Zhao apply these concepts to Outage Detection in Power Distribution Networks. By fusing nodal load forecasts with real-time line flows, they can pinpoint faults and outages even when the network's own reporting sensors might be compromised or missing.
Summary of the Intuition
The core insight of this research area is that malice has a physical cost. An attacker can change a sensor's reading, but they cannot easily change the laws of physics governing the entire system. By treating the search for compromised sensors as a combinatorial optimization problem and using formal methods (like SMT) to solve it, we can build systems that "know" when they are being lied to.