This upload contains a bias-corrected analysis and theoretical restructuring of Lyapunov-based Cognitive Control within the context of Safe Reinforcement Learning (RL). While prior work has attempted to combine Lyapunov methods with neural control architectures, it has suffered from critical theoretical and practical limitations. This paper addresses those gaps by providing a rigorous reality check on the current state of Lyapunov Cognitive Control, prioritizing scientific integrity and safety engineering realities over unvalidated performance claims.
Hashemi et al. (Thu,) studied this question.