Edge-Native, Behavior-Adaptive Drone System for Wildlife Monitoring

1 st Jenna Kline, 2 nd Rugved Katole, 3 rd Tanya Berger-Wolf, 4 th Christopher Stewart · 2025

Read Edge-Native, Behavior-Adaptive Drone System for Wildlife Monitoring by 1 st Jenna Kline, 2 nd Rugved Katole, 3 rd Tanya Berger-Wolf, 4 th Christopher Stewart (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-02285.txt.

Introduction

Drones can film wildlife up close, but flying too near stresses animals and ruins the very behavior researchers want to record. Jenna Kline and three co-authors present in 2025 a drone system designed to run on the aircraft itself and adapt its behavior to the animals' vigilance state.

Human operators face an attention bottleneck: they cannot fly the drone and monitor the whole group's stress responses at once. An edge-native system that watches vigilance and adjusts its approach automatically could gather behaviorally relevant video with less distraction to the animals. Better monitoring serves conservation science, where data quality and animal welfare must be earned together.

Catalog record

CREATOR1 st Jenna Kline, 2 nd Rugved Katole, 3 rd Tanya Berger-Wolf, 4 th Christopher Stewart
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-02285

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Edge-Native, Behavior-Adaptive Drone System for Wildlife Monitoring 1 st Jenna Kline, 2 nd Rugved Katole, 3 rd Tanya Berger-Wolf, 4 th Christopher Stewart Abstract Wildlife monitoring with drones must balance competing demands: approaching close enough to capture behaviorally-relevant video while avoiding stress responses that compromise animal welfare and data validity. Human operators face a fundamental attentional bottleneck: they cannot simultaneously control drone operations and monitor vigilance states across entire animal groups. By the time elevated vigilance becomes obvious, an adverse flee response by the animals may be unavoidable. To solve this challenge, we present an edge-native, behavior-adaptive drone system for wildlife monitoring. This configurable decision-support system augments operator expertise with automated group-level vigilance monitoring. Our system continuously tracks individual behaviors using YOLOv11m detection and YOLO-Behavior classification, aggregates vigilance states into a real-time group stress metric, and provides graduated alerts (alert vigilance → flee response) with operator-tunable thresholds for context-specific calibration. We derive service-level objectives (SLOs) from video frame rates and behavioral dynamics: to monitor 30fps video streams in real-time, our system must complete detection and classification within 33ms per frame. Our edge-native pipeline achieves 23.8ms total inference on GPU-accelerated hardware, meeting this constraint with a substantial margin. Retrospective analysis of seven wildlife monitoring missions demonstrates detection capability and quantifies the cost of reactive control: manual piloting results in 14 seconds average adverse behavior duration with 71.9% usable frames. Our analysis reveals operators could have received actionable alerts 51s before animals fled in 57% of missions. Simulating 5-second operator intervention yields a projected performance of 82.8% usable frames with 1-second adverse behavior duration, a 93% reduction compared to manual piloting. I Introduction Drones have revolutionized wildlife monitoring by enabling non-invasive observation of animal behavior in natural habitats [ 11 , 23 , 24 ] . However, successful drone-based wildlife studies require navigating a fundamental trade-off: drones must approach close enough to capture behaviorally-relevant video, but not so close that they induce stress responses that compromise both animal welfare and data validity. The vigilance escalation problem poses challenges for drone monitoring of wildlife. Prey animals exhibit natural vigilance behaviors: periodic scanning for predators that represents normal, unstressed behavior, referred to as chronic vigilance [ 14 ] . As drones approach, this can escalate: first to alert vigilance (more frequent head-up postures, alert stances), then to flight response where the entire herd flees [ 24 ] . This escalation occurs on the timescale of seconds to tens of seconds, and once flight response begins, the mission has failed: the animals are stressed, their behavior is no longer natural, and collected video is scientifically compromised [ 24 , 4 ] . Fig. 1: Edge-Native Behavior-Adaptive Drone Control for Wildlife Monitoring. The semi-autonomous drone system augments the operator’s ability to observe the wildlife and react in time to prevent adverse behavior events. The ground-control station (GCS) provides the operator with information, allowing them to adjust the automatic, behavior-adaptive navigation appropriately. The GCS recieves video stream from the drone, and in turn sends commands for where the drone should fly next, using a group-tracking navigation policy [ 19 ] , the wildlife monitor the human operator and the drone, and may flee if threatened. Manual (human-in-the-loop, HiTL) piloting faces a fundamental attentional bottleneck: a single operator cannot simultaneously control the drone, maintain safe flight operations, and continuously monitor vigilance states across many individual animals in a herd. By the time elevated vigilance becomes obvious enough for the pilot to notice, it may be too late to prevent flight response. This leads to mission failures that waste expensive field time and risk animal welfare. Our analysis of existing wildlife monitoring datasets reveals that HITL approaches result in 14 seconds average duration of adverse flight behavior per mission, with only 71.9% of collected frames usable for behavioral analysis. Some drone missions, such as mapping large areas for habitat analysis or crop monitoring, are easily automated for consistent, reliable data collection [ 28 , 3 ] . However, wildlife missions are not easily automated. Human ecological expertise is critical for successful drone missions, making fully autonomous approaches suboptimal. This is because the ecological challenges in the field extends beyond detection to behavior monitoring, and vigilance baselines vary dramatically by context [ 1 ] . Habituated animal populations near safari parks or densely populated areas tolerate closer approaches than naive populations; mothers with young exhibit elevated baseline vigilance; habitat structure, time of day, and recent disturbances all affect stress thresholds [ 24 ] . No fixed autonomous control policy can safely handle this variability, thus expert judgment based on species knowledge and field observations remains essential. We present an edge-native, behavior-adaptive drone system for terrestrial wildlife monitoring. This is a configurable decision-support system that augments human wildlife monitoring expertise with automated vigilance detection. Rather than replacing the operator, our system acts as an additional pair of eyes: computer vision models continuously monitor all individuals in the herd, quantify vigilance levels in real-time, and alert the operator before elevated vigilance escalates to unrecoverable flight response. The operator retains full control and can tune detection thresholds based on species, population, and field conditions. Just as collision avoidance systems provide preemptive warnings before impact is inevitable, our system monitors leading indicators of animal stress, i.e., elevated vigilance behaviors, allowing operators to implement graduated responses, such as reduced approach speed, pause, or retreat, before a flight response occurs. Autonomous robotics systems for measuring and minimizing disturbance to animals have been proposed in the marine domain [ 4 ] . We leverage recent advancements machine learning models capable of quickly inferring wildlife behavior to automatically detect and prevent disturbance in terrestrial wildlife [ 10 , 5 ] . We derive quantitative latency requirements from animal behavior dynamics. Real-time tracking of herd with drones has a suggested service-level-objective (SLO) of 1000ms to keep animals in frame [ 16 ] . However, behavior transitions occur almost instantly, requiring real-time monitoring of the 30 frame per second (fps) video stream, or 33 ms latency [ 14 , 7 ] . The average time from initial stress response to full flight behavior in zebra herds from seven wildlife missions was found to be 50s. However, as the duration of acute vigilance increases without intervention, so does the likelihood of permanently spooking the animals, ending the mission prematurely and spoiling the behavioral data [ 24 , 23 ] . To provide actionable warnings to operators, our system must complete detection and classification within the 33 ms. To reduce false positives, we employ a rolling 3-sample window, similar the strategy employed in Kenyan Animal Behavior Recognition study [ 10 , 14 ] , so warnings are not triggered erroneously. We demonstrate that our edge-native pipeline, which combines YOLOv11m detection (4.7 ms) with YOLO-Behavior classification (19.1 ms), achieves 23.8 ms total latency on GPU-accelerated edge devices, meeting this requirement with substantial margin. We evaluate our system using seven wildlife monitoring missions from established wildlife drone datasets (KABR [ 10 ] , WildWing [ 19 ] , and MMLA [ 17 ] ) spanning zebras, giraffes, and wild horses across diverse habitats. While retrospective analysis cannot validate closed-loop control performance, it establishes three critical system capabilities: (1) detection models achieve sufficient accuracy on

Read the complete text