Identifying safety–critical concerns in unmanned aerial vehicle software platforms with SALIENT

Abstract

Safety-related concerns may emerge during the operation of unmanned aerial vehicles (UAVs), reported by users and developers in the form of issue reports and pull requests. To help UAV developers identify safety-related concerns, we propose SALIENT, a machine learning (ML)-enabled tool that analyzes individual sentences composing the issue reports and automatically recognizes those describing a safety-related concern. The assessment of the classification performance of the tool on the issues of popular open-source UAV-related projects demonstrates that SALIENT represents a viable solution to assist developers in timely identifying and triaging safety–critical UAV issues, outperforming baselines based on ChatGPT and Google’s Bard.

Keywords

• Unmanned aerial vehicles
• Issue management
• Safety issues
• Machine learning
• Empirical study