RVReplan: A Runtime Verification Framework for Adaptive Replanning

Classical planning systems rely on action models that describe the expected preconditions and effects of actions. In realistic execution settings, these models may become incomplete, outdated, or inconsistent with observed behaviour, causing plans to fail at runtime. This paper presents RVReplan, an intelligent framework for runtime adaptation of planning action models. Given a planning domain and problem specifications, RVReplan automatically generates an executable simulation environment and synthesises runtime monitors for action preconditions and postconditions. These monitors analyse execution traces to detect missing expected preconditions and effects, while additional observed effects are recorded and used to diagnose subsequent violations when they affect the execution of the current plan. When such a violation is detected, RVReplan reasons about its likely cause, updates the relevant action specification when supported by the observations, validates the consequences of the candidate update in a simulated environment, and triggers replanning when necessary. We evaluate the framework on established International Planning Competition domains using multiple planner configurations. The results analyse monitor synthesis time, replanning time, plan size, and plan stability, showing that runtime verification can support explainable and systematic adaptation of planning models during execution. The proposed framework contributes a domain-independent mechanism for improving the robustness of intelligent planning systems at runtime.