Human-In-The-Loop Oracle Learning for Simulation-Based Testing

Abstract

Ensuring safety and providing rigorous behavioral guarantees are critical for robotic systems operating in high-stakes environments such as autonomous driving. Field testing is common, but costly and risky. Simulation-based testing offers a safer and lower-cost alternative for automatically generating traces for analysis and performance assessment. An oracle is essential for evaluating each trace, assessing whether a robot behavior fulfills key criteria such as task completion, safety, efficiency, and reliability. Supervised learning for oracle learning is accurate but costly and time-consuming due to manual labeling, whereas unsupervised learning requires no labels but often sacrifices accuracy. To overcome these limitations, we propose human-in-the-loop oracle learning as a new approach to develop and refine oracles that are capable of distinguishing good from bad behaviors with reduced manual effort. We illustrate this approach through a conceptual framework for integrating human-in-the-loop learning into robotic system evaluation.