MORTAR: A Model-based Runtime Action Repair Framework for AI-enabled Cyber-Physical Systems

TL;DR

MORTAR tackles the safety of AI-enabled CPSs with black-box DRL controllers by introducing a runtime action repair framework that builds a prediction model to forecast the STL-based safety of controller actions and uses a gradient-based optimization (via a targeted BIM) to repair unsafe actions at runtime. The prediction model acts as both a monitor and a source of gradient information to guide repairs, with safety thresholds defined by $\psi_{thres}$ and a maximum safety level $\psi_{max}=\bar{\psi}+2\sigma_{\psi}$. Large-scale robotic manipulation experiments demonstrate that MORTAR improves task completion under standard and strict STL specifications while keeping average per-step overhead around $8.8$ ms, outperforming SafeExp in most scenarios. These results suggest that real-time safety augmentation for AI-CPSs is feasible without altering or retraining the underlying DRL policy, enabling broader deployment of AI in safety-critical cyber-physical systems.

Abstract

Cyber-Physical Systems (CPSs) are increasingly prevalent across various industrial and daily-life domains, with applications ranging from robotic operations to autonomous driving. With recent advancements in artificial intelligence (AI), learning-based components, especially AI controllers, have become essential in enhancing the functionality and efficiency of CPSs. However, the lack of interpretability in these AI controllers presents challenges to the safety and quality assurance of AI-enabled CPSs (AI-CPSs). Existing methods for improving the safety of AI controllers often involve neural network repair, which requires retraining with additional adversarial examples or access to detailed internal information of the neural network. Hence, these approaches have limited applicability for black-box policies, where only the inputs and outputs are accessible during operation. To overcome this, we propose MORTAR, a runtime action repair framework designed for AI-CPSs in this work. MORTAR begins by constructing a prediction model that forecasts the quality of actions proposed by the AI controller. If an unsafe action is detected, MORTAR then initiates a repair process to correct it. The generation of repaired actions is achieved through an optimization process guided by the safety estimates from the prediction model. We evaluate the effectiveness of MORTAR across various CPS tasks and AI controllers. The results demonstrate that MORTAR can efficiently improve task completion rates of AI controllers under specified safety specifications. Meanwhile, it also maintains minimal computational overhead, ensuring real-time operation of the AI-CPSs.

Figures (4)

• Figure 1: Overview of MORTAR
• Figure 2: Workflow for a CPS with an AI controller
• Figure 3: Runtime workflow of MORTAR
• Figure 4: Five robotic manipulation tasks simulated with NVIDIA Omniverse Isaac Sim.