Development of a robotic-assisted monitoring system for nocturnal insectivores in urban parks

Thesis Abstract

Urban parks increasingly host nocturnal insectivores whose ecological roles are pivotal for pest regulation and biodiversity maintenance, yet their activity is difficult to monitor due to low light, variable terrain, and limited human access. This study addresses the gap in scalable, minimally invasive monitoring methods by developing a robotic-assisted system that integrates autonomous navigation, acoustic sensors, and environmental sensors to enhance detection, identification, and activity quantification of nocturnal insectivores in urban park contexts. The aim is to design, implement, and evaluate a portable robotic platform capable of continuous nocturnal monitoring, with objectives (1) to design a modular robotic platform equipped with high-sensitivity ultrasonic microphones, infrared thermal imaging, and an acoustical feature extraction pipeline; (2) to implement an autonomous navigation and obstacle avoidance system suitable for heterogeneous park terrains using SLAM (Simultaneous Localization and Mapping) and LiDAR fusion; (3) to develop a real-time species identification and activity estimation framework leveraging machine learning and bioacoustic analysis; (4) to assess system performance in terms of detection accuracy, operational reliability, and ecological validity against traditional survey methods; and (5) to evaluate the impact of urban park microhabitat characteristics on nocturnal insectivores’ detectability and activity patterns. The research adopts a design–implementation–evaluation approach underpinned by the theoretical frameworks of bioacoustics theory and the ecological niche theory, complemented by activity-budget and optimal foraging models to guide interpretation of activity signals. The study employs a mixed-methods design, combining quantitative sensor data with qualitative expert validation of species identifications. The population comprises nocturnal insectivores commonly observed in temperate urban parks, including bat species and nocturnal passerines, with a purposive sample of five parks selected across a metropolitan area to ensure habitat heterogeneity. A two-stage sampling strategy is used (i) a purposive selection of parks representing varied anthropogenic disturbance and green cover, and (ii) stratified nightly sampling within each park across different weather conditions over a three-month pilot, yielding an estimated 600 monitoring hours and 1,800 acoustic recordings. Data collection instruments include (i) an autonomous ground rover equipped with a 4K color camera, FLIR Boson thermal camera, an ultrasonic microphone array, a compact LiDAR sensor, a weather station, and a microcontroller-based data logger; (ii) high-resolution audio recorders and ambient environmental sensors; and (iii) a cloud-based data pipeline for feature extraction. Data analysis employs a multi-tiered approach acoustic signal processing using MFCC and spectrogram-based deep learning classifiers (CNN-LSTM) for species-level detection; environmental feature modeling using generalized linear mixed models (GLMMs) to relate detectability with temperature, humidity, noise, and vegetation structure; SLAM-based localization accuracy assessment; and validation against conventional acoustic surveys and bat detectors. Hypotheses testable within this framework include H1—robotic-assisted monitoring yields higher detection rates of nocturnal insectivores than traditional methods under identical conditions; H2—increased habitat complexity reduces detection latency; and H3—temporal activity correlates with ambient temperature and moon phase as predicted by activity-budget models. Expected findings anticipate robust operational performance of the robotic platform with mean autonomous navigation success above 92% and species-level classification accuracy exceeding 85% under controlled conditions, with a broader ecological insight that microhabitat features such as canopy cover and underbrush density significantly influence detectability and estimated activity indices. The study contributes to knowledge by delivering a validated, scalable monitoring tool for nocturnal insectivores in urban settings, advancing methods in bioacoustic sensing, robotic navigation in complex outdoors, and integrative ecological analytics that link technology-driven observations to habitat characteristics. The conclusions are anticipated to advocate for hybrid monitoring protocols combining robotic systems with citizen-science inputs and to outline design recommendations for urban park management aimed at minimizing disturbance while maximizing ecological monitoring fidelity. Recommendations include improvements to sensor fusion algorithms, deployment guidelines for year-round monitoring, and a framework for integrating robotic monitoring data into urban biodiversity dashboards and conservation planning.