Embedded Files
DR3: Machine Learning & Clustering to Detect and Filter Human-Drone Interaction Failures
DR3: Machine Learning & Clustering to Detect and Filter Human-Drone Interaction Failures
Students: Brett Porter, Jonathan Mimnaugh
Research mentor: Shahbaz Ali Khan
Human error is a significant factor in major incidents involving small Uncrewed Aerial Systems (sUAS). Fuzz testing is a software testing method aimed at detecting vulnerabilities in code; however, applying it to sUAS is dangerous and likely to produce dangerous flights and physical damage. Therefore, HIFuzz testing is conducted across three levels: a fully simulated environment with proxy human agents (L1), a simulated environment with real humans (L2), and a real-world environment with real humans (L3). Level 1 is best suited for running thousands, or even hundreds of thousands, of test cases within a reasonable timeframe due to its simulated environment and simulated human interactions. In contrast, Level 2, while still a simulated environment, involves human testing, which underscores the importance of limiting the number of cases. Transitioning from L1 to L2 through a gateway (G1) analyzes test-case results by using the K-Means clustering algorithm to identify groups of similar failures. Without effective clustering, critical types of failures may be missed and not passed down the testing pipeline to L2. The work involved enhanced the G1 gateway by creating tighter, more separated clusters using the Spherical K-Means clustering algorithm. Additionally, the work further defined the criteria to identify failure cases, so that successful outcomes, in which an sUAS completed the mission as expected, could be removed from the results. An interactive application was developed to implement this approach, enabling the user to query the data and to obtain immediate results. The application also dynamically generated visualizations of the cluster data and reported the unique features of each cluster. By improving the cluster techniques of the HIFuzz system and allowing the humans-on-the-loop to better analyze the data, the work plays a pivotal role in identifying the common factors that cause human machine interaction errors in sUAS systems.
[1] Chambers, Theodore, Michael Vierhauser, Ankit Agrawal, Michael Murphy, Jason Matthew Brauer, Salil Purandare, Myra B. Cohen, and Jane Cleland-Huang. "HIFuzz: Human Interaction Fuzzing for Small Unmanned Aerial Vehicles." In Proceedings of the CHI Conference on Human Factors in Computing Systems, pp. 1-14. 2024.
Shahbaz Ali Khan is a Master's student in Computer Science and Engineering at the University of Notre Dame, with a specialization in AI, Machine Learning, and Cloud Computing. Before that, he obtained his B.S. degree from the Siddaganga Institute of Technology majoring in Computer and Electronics. Additionally, he has hands-on experience as Senior Data Engineering Analyst at Accenture. His expertise lies in designing and implementing scalable data pipelines, optimizing database queries, and automating data ingestion, resulting in significant improvements in system performance and efficiency.
Dr Jane Cleland-Huang is the lead researcher on the Drone Response project — a system for managing and monitoring the flights of semi-autonomous small Unmanned Aerial Systems (sUAS). As part of this project, she is involved in Smart and Connected Communities (SCC) research and is working closely with the South Bend Fire Department to co-design a system in which sUAS serve as full-mission partners for emergency response scenarios.
Page updated
Report abuse