Cognitive-Flexible Control via Latent Model Reorganization with Predictive Safety Guarantees

TL;DR

This work addresses safety-critical control for partially observable systems under distributional shift by enabling cognitive-flexible online reorganization of latent beliefs. The CF--DeepSSSM framework integrates a deep stochastic state-space model with surprise-driven adaptation and a belief-space Bayesian MPC that tightens constraints adaptively to preserve safety. Theoretical guarantees include bounded posterior drift, recursive feasibility, and closed-loop stability, demonstrated through simulations under abrupt dynamics changes and observation drift. The results show that regulated latent reorganization can deliver rapid performance recovery without compromising safety, highlighting a practical approach to learning-enabled control in nonstationary cyber-physical systems.

Abstract

Learning-enabled control systems must maintain safety when system dynamics and sensing conditions change abruptly. Although stochastic latent-state models enable uncertainty-aware control, most existing approaches rely on fixed internal representations and can degrade significantly under distributional shift. This letter proposes a \emph{cognitive-flexible control} framework in which latent belief representations adapt online, while the control law remains explicit and safety-certified. We introduce a Cognitive-Flexible Deep Stochastic State-Space Model (CF--DeepSSSM) that reorganizes latent representations subject to a bounded \emph{Cognitive Flexibility Index} (CFI), and embeds the adapted model within a Bayesian model predictive control (MPC) scheme. We establish guarantees on bounded posterior drift, recursive feasibility, and closed-loop stability. Simulation results under abrupt changes in system dynamics and observations demonstrate safe representation adaptation with rapid performance recovery, highlighting the benefits of learning-enabled, rather than learning-based, control for nonstationary cyber-physical systems.

Figures (3)

• Figure 1: System overview of CF--DeepSSSM control. Observations $o_t$ are encoded into a latent belief $z_t$ with uncertainty $\Sigma_t$. A BMPC computes safe controls $u_t$, while a cognitive-flexible adaptation module updates model parameters $\theta_t$ and policy $\pi_t$ based on prediction errors under bounded reorganization.
• Figure 2: Scenario \ref{['sec:scenario1']}— Closed-loop response to an abrupt dynamics shift at $t{=}300$ (vertical dashed line). The sudden model mismatch induces a spike in the surprise signal $\mathcal{S}_t$, triggering bounded latent adaptation and a localized increase in the Cognitive Flexibility Index (CFI). Despite the representation reorganization, BMPC preserves recursive feasibility and constraint satisfaction, consistent with the bounded posterior drift guarantee (Theorem \ref{['thm:drift']}) and the constraint-tightening result (Lemma \ref{['lem:tightening']}).
• Figure 3: Scenario \ref{['sec:scenario2']} — Observation drift after $t=300$ (shaded region). The surprise signal $\mathcal{S}_t$ activates bounded reorganization of the observation model, resulting in controlled adaptation of the latent belief representation as quantified by the learning response, while BMPC preserves recursive feasibility and constraint satisfaction in accordance with Theorem \ref{['thm:drift']} and Lemma \ref{['lem:tightening']}.

Theorems & Definitions (11)

• Definition 1: Bounded posterior drift
• Theorem 1: Bounded posterior drift
• proof
• Theorem 2: Recursive feasibility
• proof
• Theorem 3: ISS under cognitive-flexible adaptation
• proof
• Corollary 1: Safety preservation
• proof
• Lemma 1: Tightening dominates prediction mismatch