Rule-based Shielding for Partially Observable Monte-Carlo Planning

TL;DR

This paper introduces a shielding method for POMCP that uses logical formulas derived from expert knowledge to identify and prevent unexpected actions, improving policy reliability and interpretability in POMDPs.

Contribution

It proposes a novel SMT-based approach to detect and shield unexpected actions in POMCP, enhancing policy safety and interpretability without sacrificing performance.

Findings

Shielded POMCP outperforms standard POMCP in benchmark tests.

The approach maintains effectiveness even with some incorrect logical parameters.

Improves policy interpretability and verification in POMDPs.

Abstract

Partially Observable Monte-Carlo Planning (POMCP) is a powerful online algorithm able to generate approximate policies for large Partially Observable Markov Decision Processes. The online nature of this method supports scalability by avoiding complete policy representation. The lack of an explicit representation however hinders policy interpretability and makes policy verification very complex. In this work, we propose two contributions. The first is a method for identifying unexpected actions selected by POMCP with respect to expert prior knowledge of the task. The second is a shielding approach that prevents POMCP from selecting unexpected actions. The first method is based on Satisfiability Modulo Theory (SMT). It inspects traces (i.e., sequences of belief-action-observation triplets) generated by POMCP to compute the parameters of logical formulas about policy properties defined by the expert. The second contribution is a module that uses online the logical formulas to identify anomalous actions selected by POMCP and substitutes those actions with actions that satisfy the logical formulas fulfilling expert knowledge. We evaluate our approach on Tiger, a standard benchmark for POMDPs, and a real-world problem related to velocity regulation in mobile robot navigation. Results show that the shielded POMCP outperforms the standard POMCP in a case study in which a wrong parameter of POMCP makes it select wrong actions from time to time. Moreover, we show that the approach keeps good performance also if the parameters of the logical formula are optimized using trajectories containing some wrong actions.