H. Kwon and J. Ueda, “Classification and Compensation of False Data Injection and Transformation Attacks on Robotic Manipulators,” in IEEE/ASME Transactions on Mechatronics, doi: 10.1109/TMECH.2026.3724374
Abstract:
Cyber-physical systems are susceptible to false data injection and transformation attacks (FDITAs) that have catastrophic consequences for the systems and their users. Such attacks are particularly dangerous as they enable various attack capabilities and opportunities. This article addresses defense against FDITAs that target either control commands or sensor observables. Standard disturbance observer (DOB) methods can detect both attack types through velocity residuals. However, DOB methods fail to compensate observable attacks in steady-state tracking because the controller continues to compute control commands from corrupted state feedback. The choice of effective adaptive compensation against FDITAs is dependent on the attack location. Control command attacks can be mitigated by preprocessing, while observable attacks require state reconstruction before they are used for command generation. A novel four-tier hybrid defense framework is proposed that classifies attack placement based on the nonzero convergence of the estimated additive bias. The method utilizes lower-diagonal-upper factorization to guarantee invertibility of estimated attack parameters. Simulation results demonstrate that the proposed method achieves up to five times reduction in tracking error compared to the DOB through location-appropriate compensation strategies. The four-tier framework is demonstrated on a 6-degree-of-freedom industrial manipulator under various FDITAs.