BEGIN:VCALENDAR VERSION:2.0 X-WR-CALNAME:EventsCalendar PRODID:-//hacksw/handcal//NONSGML v1.0//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York LAST-MODIFIED:20240422T053451Z TZURL:https://www.tzurl.org/zoneinfo-outlook/America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZNAME:EDT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZNAME:EST TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT CATEGORIES:College of Engineering,Graduate Studies,Lectures and Seminars,Th esis/Dissertations DESCRIPTION:Faculty Supervisor: Dr. Gokhan Kul, University of Massachusetts Dartmouth Committee Members:  Dr. Joshua Carberry, University of Massach usetts DartmouthDr. Yuchou Chang, University of Massachusetts Dartmouth Ab stract: Unmanned aerial vehicles or UAVs are becoming frequently used in t he military and in law enforcement applications. For the military, they no t only provide additional surveillance but aid in recon, combat, and prote ction. Drones in law enforcement are considered force multipliers by organ izations like the Federal Law Enforcement Training Center, and that gives officers multifunctional tools that can assist in daily duties. However, i t should be noted that when UAVs fly, the environment they are in can be u npredictable. UAVs are vulnerable to the environment and to autonomous pat h determination attacks which can lead to deviation from its path or crash ing. While many anomaly detection methods exist, a significant portion rel ies on limited amount of raw data, more so does not account for the physic s of the world and the relationships between that and the drone’s physic al sensors.  The method presented in this paper is an anomaly detection f ramework that uses a reinforcement learning (RL) deep Q-network (DQN) to l earn from real flight data to find normal and anomalous behaviors. In this paper, we will compare the effectiveness of using raw data and sensor fus ed data to train the RL. The main contribution of this paper to the existi ng research is the various sensor fusions created to detect malfunctions a nd anomalies of the physical sensors. Sensor-fused data involves cross-ver ifying data through sensor v sensor checks, sensor v physics checks and ph ysics v physics. To further elaborate, sensor v sensor fusions involve com paring two values to each other while sensor v physics fusions compare a s ensor value to a mathematical computation using the data. Physics v physic s sensor fusions involve comparing two laws of physics to each other. Unli ke many existing machine learning (ML) solutions which rely on raw dataset s, the solution presented compares this method to normalized sensor fused data based on drone-specific aerodynamics before evaluation. After running evaluations, we found that the sensor fused (and normalized) model consis tently achieved higher rewards during training compared to the raw data. T he sensor fused model was also superior when it came to anomaly detection. For the rewards for the DQN, the reward total for the sensor fused data w as over two times more than the raw data (32605 vs 74060). Furthermore, th ere was no case in which the accuracy score or the F1 score was higher for the baseline raw data than it was for the sensor fused. The lowest accura cy for the sensor fused was 82.78% while for raw it was 67.07%. What was s ignificant, however, was seeing that the combo fused data performed poorly in comparison to the sensor fused, and in some cases worse than the raw d ata. For the F1 scores, there was no case in which the combo fused had det ected any true anomalies leading to an average of 0% across the board for both testing datasets. This research has applications in military defense, law enforcement, and commercial uses. Its main purpose is malfunction det ection, so it’s useful for anyone who needs highly secure, tamper-proof autonomous navigation. The goal for this research is the eventual integrat ion of this framework into UAVs so it can be used in real-time. The main g oal is to integrate multi-UAV communication networks such as blockchain sm art contracts where drones can monitor each other and tell operators about potential malfunctions before the drone crashes.  For further informatio n please contact Dr Gokhan Kul at gkul@umassd.edu. \nEvent page: https:// www.umassd.edu/events/cms/finding-malfunctions-in-uavs-physical-sensors-us ing-reinforcement-learning-sensor-fusion-v-raw-data.php\nEvent link: https ://teams.microsoft.com/meet/2719811093827?p=KtSBWMxZOo55nt9oIu X-ALT-DESC;FMTTYPE=text/html:

Faculty Supervisor:
Dr. G okhan Kul\, University of Massachusetts Dartmouth

\n

Committee Member s: 
Dr. Joshua Carberry\, University of Massachusetts Dartmouth
Dr. Yuchou Chang\, University of Massachusetts Dartmouth

\n

Abstrac t:

\n

Unmanned aerial vehicles or UAVs are becoming frequently used i n the military and in law enforcement applications. For the military\, the y not only provide additional surveillance but aid in recon\, combat\, and protection. Drones in law enforcement are considered force multipliers by organizations like the Federal Law Enforcement Training Center\, and that gives officers multifunctional tools that can assist in daily duties. How ever\, it should be noted that when UAVs fly\, the environment they are in can be unpredictable. UAVs are vulnerable to the environment and to auton omous path determination attacks which can lead to deviation from its path or crashing. While many anomaly detection methods exist\, a significant p ortion relies on limited amount of raw data\, more so does not account for the physics of the world and the relationships between that and the drone ’s physical sensors.  The method presented in this paper is an anomaly detection framework that uses a reinforcement learning (RL) deep Q-network (DQN) to learn from real flight data to find normal and anomalous behavio rs. In this paper\, we will compare the effectiveness of using raw data an d sensor fused data to train the RL. The main contribution of this paper t o the existing research is the various sensor fusions created to detect ma lfunctions and anomalies of the physical sensors. Sensor-fused data involv es cross-verifying data through sensor v sensor checks\, sensor v physics checks and physics v physics. To further elaborate\, sensor v sensor fusio ns involve comparing two values to each other while sensor v physics fusio ns compare a sensor value to a mathematical computation using the data. Ph ysics v physics sensor fusions involve comparing two laws of physics to ea ch other. Unlike many existing machine learning (ML) solutions which rely on raw datasets\, the solution presented compares this method to normalize d sensor fused data based on drone-specific aerodynamics before evaluation . After running evaluations\, we found that the sensor fused (and normaliz ed) model consistently achieved higher rewards during training compared to the raw data. The sensor fused model was also superior when it came to an omaly detection. For the rewards for the DQN\, the reward total for the se nsor fused data was over two times more than the raw data (32605 vs 74060) . Furthermore\, there was no case in which the accuracy score or the F1 sc ore was higher for the baseline raw data than it was for the sensor fused. The lowest accuracy for the sensor fused was 82.78% while for raw it was 67.07%. What was significant\, however\, was seeing that the combo fused d ata performed poorly in comparison to the sensor fused\, and in some cases worse than the raw data. For the F1 scores\, there was no case in which t he combo fused had detected any true anomalies leading to an average of 0% across the board for both testing datasets. This research has application s in military defense\, law enforcement\, and commercial uses. Its main pu rpose is malfunction detection\, so it’s useful for anyone who needs hig hly secure\, tamper-proof autonomous navigation. The goal for this researc h is the eventual integration of this framework into UAVs so it can be use d in real-time. The main goal is to integrate multi-UAV communication netw orks such as blockchain smart contracts where drones can monitor each othe r and tell operators about potential malfunctions before the drone crashes . 

\n

For further information please contact Dr Gokhan Kul at gkul@umassd.edu. 

Event page: https: //www.umassd.edu/events/cms/finding-malfunctions-in-uavs-physical-sensors- using-reinforcement-learning-sensor-fusion-v-raw-data.php
Event lin k: https://teams.microsoft.com/meet/2719811093827?p=KtSBWMxZOo55nt9oIu

DTSTAMP:20260421T151434 DTSTART;TZID=America/New_York:20260507T093000 DTEND;TZID=America/New_York:20260507T103000 LOCATION:Dion 311 and Teams (https://teams.microsoft.com/meet/2719811093827 ?p=KtSBWMxZOo55nt9oIu) SUMMARY;LANGUAGE=en-us:Finding Malfunctions in UAV’s Physical Sensors usi ng Reinforcement Learning; Sensor Fusion v. Raw Data UID:c714037da2324c28b2c10423c5f7c2ab@www.umassd.edu END:VEVENT END:VCALENDAR