National Science Foundation, CPS: Small: Enhancing Cybersecurity in Remote-Controlled Robots Using Depth-Compressed Encrypted Adaptive Control and Virtual Trip-Wire Overflow Traps (#2448402), PI: Ueda, $600,000, 7/15/2026-7/14/2029
This project develops a unified defense framework against false data injection attacks in networked cyber-physical systems by integrating two complementary mechanisms: a depth-compressed encrypted adaptive controller and a probabilistic virtual overflow trap, both realized through somewhat homomorphic encryption, a cryptographic technique permitting arithmetic computations directly on encrypted data without decryption. The central challenge is that encrypting adaptive control algorithms causes rapid accumulation of computational noise with each arithmetic operation, eventually causing decryption failure known as overflow. The first objective resolves this by restructuring the Model Reference Adaptive Control law through algebraic circuit rewrite rules that halve the algorithm’s multiplicative depth, making real-time encrypted control feasible. The second objective converts this managed noise budget into a security asset: legitimate operation remains below the overflow threshold, while adversarial signal manipulations of sufficient complexity inevitably exceed it, triggering attack detection with probability converging to one within a relatively small number of control cycles. A three-node international experimental testbed provides empirical validation and will be shared with the research community upon project completion.