Revealing Safety-Critical Scenarios for UTM via Transformer

2026-06-30

AI Summary

This work addresses the critical gap in existing Unmanned Traffic Management (UTM) systems, which lack effective mechanisms to expose safety-critical vulnerabilities and suffer from ambiguous reward signals during testing. To tackle this challenge, the study introduces a risk-oriented reinforcement learning framework that incorporates Transformers for the first time in UTM safety validation. By leveraging attention mechanisms to model state dependencies and integrating a policy network with a domain-constrained action sampler, the approach generates targeted test scenarios that effectively circumvent system self-healing behaviors responsible for long-tail failures. Evaluated over 700 hours of simulation, the method demonstrates an eightfold improvement in vulnerability discovery efficiency compared to expert-guided testing and successfully uncovers multiple critical edge cases missed by conventional approaches.

Abstract

Unmanned Traffic Management (UTM) systems are cloud-based platforms designed to manage and coordinate multiple aerial vehicles remotely. UTM systems are safety-critical which cannot tolerate failures like crash or collision. To reveal latent vulnerabilities, there are neither optimal failure-exposing demonstrations nor clear reward signals. Additionally, UTM's self-healing capability introduces the ``long-tail effect'' of critical failures. We propose framing UTM vulnerability discovery as a sequence modeling problem amenable to transformer-based RL architectures. Our approach leverages attention mechanisms to directly model the relationship among system states, and predict optimal actions. Our framework introduces a Policy Model that generates targeted test scenarios and an Action Sampler that enforces domain constraints. We use a risk-based reward function to guide exploration. Through extensive evaluation on a 700-hour simulation study, we demonstrate an 8$\times$ improvement in vulnerability discovery efficiency compared to expert-guided testing. It also discovers critical edge cases that traditional methods have missed.

Problem

• UTM
• safety-critical scenarios
• vulnerability discovery
• long-tail effect
• failure exposure

Innovation

• Transformer-based RL
• sequence modeling
• safety-critical scenarios
• vulnerability discovery
• Unmanned Traffic Management (UTM)