INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
AquaNova: An AI-Powered, CNN-Based Aquatic Trash Collector  
with Whale-Inspired Suction, Solar-Battery Operation, Using  
Arduino, Waypoint Navigation, and Dijkstra’s Algorithm  
John Ivan P. Ello1, Charity Joy Ladia2, Ryan Paul Obligar3  
College of Information Systems and Technology Management, Pamantasan ng Lungsod ng Maynila,  
Philippines  
DOI: https://doi.org/10.51584/IJRIAS.2025.10100000173  
Received: 10 November 2025; Accepted: 17 November 2025; Published: 20 November 2025  
ABSTRACT  
Mismanaged solid waste, especially plastic, remains a major ecological threat, with the Philippines  
contributing heavily to marine pollution due to its sachet-driven consumption and highly contaminated river  
systems. This study responds to the need for an automated solution that reduces environmental and health risks  
caused by floating waste.  
The research developed and evaluated AquaNova, an autonomous, solar-powered floating robot equipped with  
an AI vision system, GPS waypoint navigation, and a whale-inspired suction mechanism for efficient waste  
detection and collection. Built using the Prototype SDLC, the system integrates CNN-based detection through  
YOLOv8 and OpenCV on a Raspberry Pi, with performance assessed through quantitative metrics on detection  
accuracy, navigation, and operational reliability.  
The AI detection model achieved 80 percent overall accuracy, including 100 percent precision for plastic and  
correct identification of non-trash items in 8 out of 10 cases. Reliability tests showed 92.6 percent system  
availability, confirming stable operation and dependable GPS waypoint navigation.  
AquaNova’s successful implementation demonstrates the feasibility of autonomous, sustainable aquatic waste  
cleanup. By combining accurate AI-based detection with reliable navigation, it reduces waste accumulation  
and helps prevent blockages. Future improvements should enhance metal detection performance and add night  
vision to support continuous operation.  
Keywords:Trash Collection, Autonomous Robot, AI Detection, Water Pollution Robot, Autonomous  
Navigation  
INTRODUCTION  
Mismanaged solid waste, especially plastic, remains a major ecological burden in the Philippines, where  
millions of tons of waste are generated annually and an estimated 20 percent enters the ocean. With over 7,600  
islands and a heavy dependence on single-use sachets, the country contributes significantly to global marine  
pollution. These plastics break down into microplastics that threaten marine life, while clogged waterways and  
outdated drainage systems worsen flooding and increase health risks. Other waste types such as food, paper,  
metal, clothing, rubber, and glass also harm ecosystems. Studies show that more than a quarter of the world’s  
top plastic-emitting rivers are in the Philippines, including the Pasig River and 18 others among the top 50  
most polluting globally.  
AquaNova was developed as an autonomous solution to address this waste problem. It detects and collects  
floating solid waste using Arduino Nano–based control, AI-powered CNN detection, and a dual power system  
combining a 40W solar panel with lithium-ion batteries. Its vision system, powered by a Raspberry Pi running  
YOLOv8, OpenCV, and CNN models, identifies debris in real time. Inspired by whale-like movement and a  
mouth-style intake mechanism, AquaNova glides across water surfaces and gathers trash efficiently.  
Page 1981  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
The robot navigates via the NEO-M8N GPS module and GY-271 HMC5883L compass, with AI-driven  
adjustments when trash is detected. Movement is supported by dual propellers, and waste is collected through  
a submerged whale-inspired chamber whose lifting “lip” secures debris inside two separate internal chambers.  
AquaNova operates in three modes: Cycling Waypoint Mode for continuous GPS-based patrolling, Trash  
Collection Mode for automatic deviation when trash is detected, and Trash No Waypoint Mode for manual or  
IR-sensor-triggered collection. The IR sensor also determines when the chambers are full and directs the robot  
back to its home waypoint. The Raspberry Pi handles AI processing and sends detection data, including trash  
type and location, to the Arduino for navigation.  
Its range is extended through LoRa technology for long-distance control and telemetry. A monitoring device  
displays real-time data such as battery voltage, IR alerts, propeller status, servo angle, heading, GPS  
coordinates, satellite count, HDOP, and mouth activity. The AquaNova Control software enables manual  
adjustments to modes, throttle, steering, and intake functions.  
For optimal performance, AquaNova requires open-field operation for accurate GPS signals and sufficient  
lighting for visual detection. It performs best within 9V to 12.6V, with peak efficiency at 12.6V. Navigation is  
guided by a waypoint algorithm supported by Dijkstra’s Algorithm for efficient targeting. A Windows  
interface provides real-time video, navigation data, and detection logs for operator monitoring.  
With AI-driven detection, a whale-inspired intake design, dual-chamber segregation, precise GPS-compass  
navigation, and long-range communication, AquaNova presents an autonomous and eco-efficient solution for  
mitigating floating solid waste in Philippine waterways. Its development supports ongoing advancements in  
the HydroSent project, enhancing waste collection efficiency, detection accuracy, and navigation performance  
for sustainable water pollution management.  
Statement of the Problem  
Through an assessment of floating waste and its impact on water pollution, the researchers identified critical  
challenges in waste management. This study aims to address the following issue:  
How can an autonomous system ensure waste collection without interfering with freshwater species?  
Objective of the Study  
To implement an AI-powered detection system using YOLOv8, OpenCV, and CNN on a Raspberry Pi for real-  
time recognition of floating solid waste, ensuring accurate trash detection while preserving freshwater species.  
LITERATURE REVIEW  
A. Review of Related Literature  
The Philippines is the world’s third-largest ocean plastic polluter, discarding about 2.7 million tons of plastic  
annually. Daily waste generation further highlights the scale of the problem, with over 163 million sachets, 48  
million shopping bags, and 45 million thin-film plastic bags disposed of each day (Gorecho, 2024). Cleanup  
efforts reflect this severity: the International Coastal Cleanup collected 352,479 kilograms of trash across 250  
coastal sites, dominated by plastic bags, food wrappers, bottles, and sachets (Abreo & Kobayashi, 2024). Even  
localized initiatives, such as Romblon’s floating trash traps made from recycled bottles, gathered 285.76 kg of  
waste composed of 68 percent biodegradable, 14 percent recyclable, 12 percent residual, and 6 percent special  
waste (Gacu, 2023). These figures emphasize the need for scalable, automated, and energy-efficient cleanup  
solutions.  
To address this need, the Department of Science and Technology – Metals Industry Research and  
Development Center developed a barge-type garbage collector capable of removing large quantities of floating  
debris in Metro Manila’s waterways. Although effective on a large scale, its fixed-path, non-autonomous  
design limits responsiveness to shifting waste patterns. AquaNova overcomes this by combining YOLOv8-  
Page 1982  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
based AI detection, Convolutional Neural Networks, and Arduino-controlled autonomous navigation, enabling  
real-time waste recognition and dynamic path adjustments.  
A related development is the solid waste filtering robot by Lecitona, Gamboa, Songco, and Abuan (2020),  
which filtered contaminated water, detected full capacity through sensors, and transmitted data wirelessly.  
While demonstrating autonomous waste filtration in shallow waters, its design focused on water purification  
rather than targeted solid waste retrieval. AquaNova extends this concept through targeted suction, AI-powered  
detection, Dijkstra’s Algorithm for optimized pathfinding, and onboard waste segregation, allowing operation  
in broader and more variable aquatic environments.  
Together, these technologies show the shift from manual and static systems toward intelligent, sensor-driven  
aquatic waste management. AquaNova integrates these advancements into a fully autonomous, energy-  
efficient platform capable of real-time detection, adaptive navigation, and efficient waste segregation,  
supporting sustainable environmental management efforts locally and globally.  
B. Review of Related Studies  
The increasing challenge of aquatic waste pollution has led researchers to develop advanced, technology-  
driven approaches for cleaner waterways. From mechanical collectors to AI-powered autonomous systems,  
recent studies highlight how robotics, microcontrollers, and machine learning can improve waste detection,  
collection, and overall environmental response. These works form the technological foundation of AquaNova  
by demonstrating innovations in control systems, navigation, energy management, object recognition, and  
sustainable aquatic design.  
To address waterway pollution in Metro Manila, the Department of Science and Technology–Metals Industry  
Research and Development Center developed a barge-type garbage collector capable of removing solid waste  
and water hyacinth. Its mechanical rakes and conveyors allow large-scale debris removal, but it follows fixed  
routes and cannot adapt to shifting waste patterns. While effective for bulk collection, its lack of autonomy  
limits coverage. AquaNova builds on this by integrating YOLOv8 and CNN-based visual detection for real-  
time recognition and route adjustments, and its compact design enables operation in narrow or shallow  
waterways where large barges cannot.  
UN-Habitat Philippines (2023) introduced an AI-assisted waste mapping system using satellite imagery and  
drones to identify plastic waste hotspots. Although effective for high-level assessment, it relies on post-  
processing and human evaluation before action can be taken. AquaNova removes this delay by embedding AI  
detection directly onboard, allowing immediate identification and retrieval of floating debris and merging data  
collection with rapid cleanup capability.  
Lecitona, Gamboa, Songco, and Abuan (2020) developed a solid waste filtering robot for shallow waters that  
filtered contaminated water, separated debris, and returned cleaner water to the environment. With tactile  
sensors, wireless modules, and real-time monitoring, it demonstrated efficient localized filtration. AquaNova  
advances this model through targeted suction, YOLOv8-based visual detection, optimized navigation using  
Dijkstra’s Algorithm, and bin-level sensing for intelligent decision-making, enabling operation in wider and  
more dynamic environments.  
Overall, previous systems show strong progress in waste collection, filtration, and sensing but often lack  
adaptive navigation, onboard AI decision-making, or integrated waste segregation. AquaNova addresses these  
gaps by combining AI-based detection, optimized routing, hybrid solar-battery power, capacity monitoring,  
and dual-chamber segregation in a single autonomous platform, enhancing operational efficiency and  
supporting long-term waterway rehabilitation.  
METHODOLOGY  
The research methodology of this study ensures the systematic development and implementation of  
AquaNova, an AI-powered, whale-inspired aquatic trash collector. It follows a methodical approach that  
integrates Convolutional Neural Networks with a Raspberry Pi vision system running YOLOv8 and OpenCV  
Page 1983  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
for real-time trash detection. Navigation is guided by Dijkstra’s Algorithm, with hardware and software  
components calibrated for efficiency and reliability. This section outlines the processes for data gathering,  
processing, and analysis of detection accuracy, navigation performance, and waste collection efficiency.  
Additionally, it facilitates the evaluation of system standards through quantitative metrics and expert  
validation, ensuring its effectiveness in autonomous waste removal.  
A. Research Design  
The research design of this study revolves around the Software Development Lifecycle Model applied to  
AquaNova, an AI-powered aquatic trash collector. It follows a systematic approach that incorporates the key  
principles of the Evolutionary Prototype SDLC to ensure efficient development, testing, and deployment.  
Additionally, the researchers conducted an interview and presentation with the Department of Environment  
and Natural Resources –National Capital Region (DENR-NCR) to validate the project's environmental  
relevance and gather expert insights.  
Fig. 1 Prototype SLDC as AquaNova’s systematic Approach  
The development of AquaNova, an upgraded aquatic trash collector, follows the Prototype Software  
Development Life Cycle (SDLC). This model allows researchers to build a working prototype early in the  
development stage, test its features with stakeholders, and refine the system through iterative feedback before  
final deployment. Unlike the Agile model, which emphasizes adaptability across sprints, the Prototype model  
emphasizes early visualization, evaluation, and user involvement. This ensures that AquaNova meets both  
technical requirements and environmental objectives. As shown in Fig. 1, there are six stages: Requirement  
Analysis, Quick Design, Prototype Development, Customer Evaluation & Refinement and Final System  
Development & Deployment.  
Requirement Analysis:  
This phase involves brainstorming sessions, consultations, and data gathering to clearly define the project’s  
objectives and requirements. The researchers analyzed both the existing problems and the goals that AquaNova  
must accomplish within the given timeframe, while also addressing the shortcomings of its predecessor,  
HydroSent.  
Page 1984  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 2 List of required materials and interview conducted with DENR-NCR  
Fig. 2 presents an interview conducted by the researchers with the Department of Environment and Natural  
Resources – National Capital Region (DENR-NCR).The discussion addressed key environmental concerns,  
including the identification of the main types of trash commonly found in bodies of water, and determined the  
specific features and functionalities required to be implemented in AquaNova. In addition, the researchers  
compiled a comprehensive list of the materials needed for the system and sourced reliable links for their  
procurement, thereby finalizing the goals and objectives of this study. The inputs for this stage were  
categorized into three areas: Knowledge Inputs – Artificial Intelligence (AI) Navigation, Machine Learning  
techniques, Object Detection using YOLOv8, and Waypoint Algorithms. Hardware Inputs – Raspberry Pi,  
Arduino Nano, GPS modules, environmental sensors, a dual-chamber waste collection system, and solar  
panels. Software Inputs – OpenCV, PyTorch, Arduino IDE, and Visual Studio Code, which served as the  
primary platforms for system development and testing.  
The process involved defining the project goals, specifying the required features, and identifying system  
constraints, which formed the foundation for AquaNova’s design. The outputs of this stage were consolidated  
into a draft list of functionalities, which included the whale-inspired suction mechanism, improved navigation  
capabilities, and optimized trash collection efficiency.  
Quick Design:  
At this phase, researchers created conceptual sketches and system architecture diagrams of AquaNova. It  
involves creating the initial design of AquaNova, a crucial step in establishing the foundation for development.  
The researchers utilized GrabCAD to develop the detailed illustration and 3D model of the system, ensuring  
that the design is both functional and feasible for implementation. This stage also focuses on aligning the  
model’s features with the project’s objectives to guarantee its effectiveness in real-world application  
Fig. 3 3D Model of AquaNova  
Fig.3 showcases the quick design stage of the Evolutionary Prototyping Model for AquaNova involved the  
creation of initial conceptual sketches and system architecture diagrams that served as the blueprint for the  
prototype. This stage emphasized translating the identified requirements into preliminary visual and structural  
Page 1985  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
representations, allowing the researchers to outline the system’s essential components and their  
interconnections before physical development.  
The design primarily focused on four core aspects: the whale-inspired suction mechanism to enhance waste  
intake efficiency, the integration of solar-powered energy systems to improve sustainability and operational  
longevity, the dual-propeller maneuvering system for stability and adaptability in aquatic environments, and  
the implementation of AI-based trash detection to optimize real-time identification and collection of debris.  
These conceptual designs provided the foundation for building the initial prototype and guided subsequent  
refinement through stakeholder feedback and iterative development.  
Development:  
The development of AquaNova started with an extensive data-gathering phase, which focused on collecting  
relevant information, identifying key problems, setting clear objectives, and outlining the necessary hardware  
and software requirements. These well-defined requirements served as a strong foundation for the system’s  
creation. During the development process, the researchers conducted continuous testing to detect and resolve  
bugs, ensuring that AquaNova met the established requirements and performed efficiently in collecting and  
managing waste from aquatic environments.  
Fig. 4 Developing Arduino Code for Maneuver  
In Fig. 4, to support the functionality of the prototype, the researchers developed and integrated software using  
Visual Studio Code and the Arduino IDE. Visual Studio Code provided a scalable environment for managing  
project code, while the Arduino IDE enabled direct programming of the microcontroller for precise motor  
control and activation of detection triggers. This integration ensured communication between hardware and  
software components, allowing the prototype to maneuver in aquatic environments and respond to sensor  
inputs. Although limited in performance compared to the final system, the prototype served its purpose by  
demonstrating AquaNova’s core mechanisms and establishing a working foundation for iterative testing and  
stakeholder evaluation.  
Customer Evaluation and Refinement:  
This phase focuses on testing AquaNova for potential bugs and errors. It involves evaluating the system’s  
features and functionality to identify any gaps or unmet requirements. Through rigorous testing, the  
researchers ensure that AquaNova performs as intended, from maneuvering and detection to trash collection,  
while delivering optimal results in real-world conditions. A thorough assessment of the system’s usability,  
functional suitability, performance, and dependability is conducted to verify its effectiveness in collecting  
waste from bodies of water. Testing guarantees a well-rounded and reliable aquatic trash collection system.  
Page 1986  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 5 Testing AquaNova’s System  
The researchers conducted a seven-day reliability and performance test of AquaNova under controlled pool  
conditions, operating the robot continuously for 24 hours per day to simulate real-world waste collection  
scenarios. Alternating periods of active operation and induced interruptions were used to record downtime and  
repair intervals. Throughout the test, three key parameters were monitored: total operating time, which reached  
168 hours; total downtime, which accumulated to 24 hours; and the number of failures, which totaled seven  
incidents involving malfunctions such as navigation drift, suction blockage, or sensor inaccuracy. Each failure  
event was documented with its exact occurrence time, repair duration, and restoration time before AquaNova  
resumed normal operation.  
Using these recorded values, the researchers computed AquaNova’s reliability metrics. The Mean Time to  
Repair (MTTR) was 3.43 hours, derived from total downtime divided by failures. The Mean Time Between  
Repairs (MTBR) was 24 hours based on total operating time divided by failures. The Mean Time Between  
Failures (MTBF) was calculated as 20.57 hours using MTBF = MTBR – MTTR. System availability was  
determined using Availability = Total Operating Time ÷ (Total Operating Time + Total Down Time), resulting  
in 85.71 percent. These results demonstrate AquaNova’s dependability, endurance, and strong operational  
readiness, confirming that the system can maintain continuous functioning with minimal downtime in  
freshwater conditions.  
1)  
Deployment:  
This phase involves preparing and deploying the developed AquaNova system for its intended end-users. The  
goal is to ensure proper implementation, smooth integration with the operating environment, and efficient  
utilization by users for effective aquatic trash collection and detection  
Fig. 6 Deployment of AquaNova  
Page 1987  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 7 The System Architecture of AquaNova  
Based on Fig. 6, the AquaNova prototype and software are now fully developed and ready for deployment.  
The system, including its hardware components and detection software, has been integrated and tested,  
ensuring proper communication between the devices such as the Raspberry Pi, Arduino, and onboard sensors.  
This readiness marks the transition from development to actual operation for aquatic trash collection and  
detection.  
B. Proposed Algorithm and System Architecture  
System Architecture:  
The system architecture provides a concrete structural framework for AquaNova, showcasing its integration of  
hardware and software components to enable efficient aquatic trash collection and detection. It follows a  
layered architecture, consisting of key components such as raw input from sensors and cameras, data  
processing for object detection, navigation algorithms for maneuvering, and a user-friendly interface for  
monitoring and control.  
Fig. 7 illustrates the overall operation of AquaNova, an AI-based autonomous robot designed for trash  
detection and collection in water bodies.The process begins by checking for predefined coordinates, and if  
none exist, the user inputs waypoints for navigation. Once the waypoints are set, the robot enters Cycling  
mode, navigating through each point while continuously assessing its orientation and calculating distances.  
During movement, the Raspberry Pi camera activates YOLOv8 for object detection. Objects not recognized by  
the system are ignored, while trained classes are verified through OpenCV. Upon detecting a match,  
AquaNova diverts from its route, collects the identified waste using its lid and chamber mechanism, and then  
resumes its Cycling mode.  
After completing all waypoints, AquaNova either loops through them again or returns to its home waypoint  
when its storage is full. This workflow addresses the first three problems of the study by demonstrating the  
system’s operational procedures, AI-based detection, and waypoint navigation. Through this process,  
AquaNova effectively achieves autonomous movement and efficient waste collection in aquatic environments.  
Circuit Diagram:  
The circuit diagram illustrates the detailed electrical framework for the AquaNova system, showcasing the  
integration of power supply, control modules, and sensor components to enable efficient aquatic trash  
collection and detection. It incorporates essential elements such as a solar panel power source, motor drivers  
for propulsion control, various environmental and navigation sensors, and a Raspberry Pi for centralized  
processing. This configuration ensures continuous energy supply, precise system control, and real-time data  
acquisition, forming the backbone of AquaNova’s operational capabilities.  
Page 1988  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 8 The System Circuit Architecture of AquaNova  
Fig. 8 illustrates AquaNova’s energy and control flow, beginning with the solar panel, which converts sunlight  
into electrical energy regulated by the solar charge controller (PWM) before being stored in the battery pack  
and managed by the BMS. A pre-battery ON/OFF switch supports maintenance, while a post-battery ON/OFF  
switch controls power distribution to electronic modules. Stored energy passes through DC/DC step-down  
converters to supply stable voltages for the Arduino Nano and Raspberry Pi.  
The Arduino Nano operates as the primary microcontroller, driving the motor controller for the left and right  
propellers and handling four servo motors for steering and waste collection. It also processes inputs from the  
compass, GPS, and two infrared sensors for navigation, obstacle detection, and alignment. The Raspberry Pi  
functions as the high-level processor, using a camera to capture real-time footage for AI-based waste detection  
and classification. By coordinating visual processing and microcontroller commands, AquaNova achieves  
autonomous operation, combining renewable energy, precise control, and intelligent detection to clean aquatic  
environments efficiently.  
Page 1989  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 9 Overall System Flowchart  
Figure 9 shows the overall process of AquaNova, an AI-based autonomous robot for trash detection and  
collection in water bodies. The operation begins by checking for predefined coordinates; if none exist, the user  
inputs waypoints for navigation. Once set, the robot enters Cycling mode, navigating through each waypoint  
while continuously assessing its orientation and calculating distances. During movement, the Raspberry Pi  
camera activates YOLOv8 for object detection. Unrecognized objects are ignored, while trained classes are  
verified through OpenCV. When a match is found, the robot diverts from its route, collects the detected waste  
using its lid and chamber mechanism, and then resumes Cycling mode.  
When all waypoints are completed, AquaNova loops through them or returns to its home waypoint once its  
storage is full. This figure addresses the first three problems of the study by illustrating the system’s  
operational workflow, AI-based detection, and waypoint navigation, demonstrating how AquaNova achieves  
autonomous movement and efficient waste collection.  
Proposed Algorithms:  
This section presents the algorithms integrated into the AquaNova system to enable efficient autonomous  
navigation, accurate object detection, and effective waste collection. Each algorithm is designed to address  
specific operational requirements, such as real-time identification of floating debris, optimal path planning  
between waypoints, and adaptive maneuvering during trash retrieval. By combining AI-based detection models  
with navigation and decision-making algorithms, AquaNova ensures reliable performance in varying  
environmental conditions while maintaining precision in both movement and collection tasks.  
The following subsections describe the selected algorithms, their operational flow, and their roles in achieving  
the system’s objectives.  
Page 1990  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 10 Waypoint Navigation Algorithm  
Fig. 10 outlines the Waypoint Navigation algorithm, designed to proactively remove floating waste across  
multiple waypoints within bodies of water. The system begins by loading a list of predefined waypoints and  
determining its current location via the GPS module (NEO-6M) and HMC5883L compass module. Once the  
current position is established, it autonomously navigates to the first waypoint.  
During navigation, if trash is detected, the system halts movement, activates the trash collection process, and  
records the current waypoint. After collecting the waste, the system resumes its journey from the saved  
waypoint and continues following the planned route. If no trash is detected, it moves directly to the next  
waypoint. This process ensures that the robot proactively removes waste using proper maneuvering systems.  
Once all waypoints are visited, the robot loops through them; however, if the IR sensor detects that the  
container is full, it returns to Waypoint 1, continuing to prevent blockages and maintain smoother water flow.  
Page 1991  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 11. Dijkstra’s Algorithm  
Fig. 11 illustrates Dijkstra’s Algorithm, designed to efficiently remove floating waste by determining the  
shortest route to a detected object. The process begins when the trash detection module identifies waste in the  
water. At this point, the ongoing Cycling Waypoint Navigation sequence pauses, and the Trash Collection  
Method is activated. The system records the current waypoint to ensure it can return to its original path after  
collection. Using the GPS module, the robot determines its current coordinates and applies Dijkstra’s  
Algorithm to calculate the shortest route to the detected waste.  
After identifying the optimal path, the system analyzes orientation data from the compass module to ensure  
proper heading alignment. Once both position and orientation are confirmed, control commands are sent to the  
motors, directing the robot toward the trash. When the object enters the collection range, the system activates  
the lid mechanism, lifting it to collect the waste efficiently before resuming the waypoint navigation process.  
Page 1992  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
C. Methods and Tools  
In software development, methods and tools play a crucial role in maintaining product quality. They define the  
approach to system development, guide data processing, and shape how researchers implement procedures to  
build and refine systems. Choosing the appropriate methods and tools is essential, as it influences the  
collection of requirements, the management of modifications, and the delivery of the final prototype.  
1) Camera Detection and Monitoring:  
Fig. 12 Real-Time Video Feed and Detected Trash  
AquaNova features a real-time video feed and detected trash information accessible through its Monitoring  
Software and Web Server, as shown in Fig. 12. This system enables continuous observation of the robot’s  
surroundings and detection activity while operating in aquatic environments. The live feed allows users to  
monitor the identification of floating waste in real time, while the interface displays critical data such as trash  
type, position, and collection status.  
The system focuses on accurately detecting floating solid waste in real-time, allowing the robot to distinguish  
between debris and other objects in the water while minimizing any impact on freshwater species. The  
detection process is critical for ensuring that only relevant waste is targeted for collection, maintaining  
environmental safety.  
The AI system is run on a Raspberry Pi, which serves as the high-level processing unit. A pre-trained  
Convolutional Neural Network (CNN) interprets visual data, recognizing patterns to identify and classify  
debris accurately. YOLOv8 (Ultralytics You Only Look Once, Version 8) is integrated for high-performance  
object detection, while OpenCV manages real-time image processing, video streaming, and feature extraction  
from the camera feed to enhance detection precision. Python and PyTorch provide the framework for model  
training and neural network execution, with Pillow (PIL) supporting image manipulation during both training  
and detection.  
This setup enables AquaNova to continuously analyze captured video and determine whether a detected object  
matches a stored class, such as plastic or metal. When a match is confirmed, the robot switches to Trash  
Page 1993  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Collection Mode; otherwise, non-target objects, including freshwater species, are ignored. This approach  
ensures that AquaNova efficiently collects waste while preserving aquatic life.  
Fig. 12 Real-Time Video Feed and Detected Trash  
Fig. 13 illustrates AquaNova’s object detection process for identifying and classifying floating waste. The  
Raspberry Pi camera captures visual data, which is processed using YOLOv8 and a pre-trained CNN model.  
OpenCV handles the video stream, resizing and optimizing images for detection. When a detected object  
matches a stored class, AquaNova switches to Trash Collection Mode and pauses Cycling Waypoint Mode;  
unmatched objects are ignored, and navigation continues. This AI-based recognition ensures effective waste  
collection while minimizing harm to marine life.  
Page 1994  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 14 Monitoring Device  
Fig. 14 illustrates the operation of the monitoring device once trash is detected. Upon initialization, the device  
checks Wi-Fi connectivity, performs a system status check, and displays a live video feed. When an object is  
identified as trash, the system records key information such as trash type, box size, and confidence rate.  
Page 1995  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Confusion Matrix:  
TABLE I CONFUSION MATRIX  
Predicted Result Predicted Result Predicted Result  
Actual Result  
Actual Result  
Actual Result  
The Confusion Matrix compares the actual results with the predicted results, determining the accuracy,  
precision, and other key metrics of AquaNova’s trash detection model. This evaluation is conducted after the  
AI-based object detection process, enabling the researchers to assess the performance of the system’s  
classification capabilities in identifying aquatic waste.  
The following statistical tools were used in this study to examine the data gathered by the developers for this  
study:  
Accuracy (Confusion Matrix) – Used to measure the correctness of AquaNova’s AI-powered detection system  
in classifying floating solid waste, based on the comparison between the system’s predicted results and the  
actual observed outcomes during testing.  
Formula:  
Fig. 15 Accuracy (Confusion Matrix)  
Where:  
TP = True Positives  
TN = True Negatives  
FP = False Positives  
FN = False Negatives  
Precision (Confusion Matrix) - Measures the accuracy of positive predictions.  
Formula:  
Fig .16 Precision (Confusion Matrix)  
Where:  
TP = True Positives  
TN = True Negatives  
FP = False Positives  
Page 1996  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
FN = False Negatives  
Recall (Confusion Matrix) - Measures how well the model identifies actual positives.  
Formula:  
Fig .17 Recall (Confusion Matrix)  
Where:  
TP = True Positives  
TN = True Negatives  
FP = False Positives  
FN = False Negatives  
F1-Score (Confusion Matrix) - Balances precision and recall.  
Formula:  
Fig. 18 F1 Score (Confusion Matrix)  
Where:  
TP = True Positives  
TN = True Negatives  
FP = False Positives  
FN = False Negatives  
Test Scripts:  
Table Ii Camera Detection Functionality Test  
Page 1997  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
RESULTS AND DISCUSSION  
In this chapter, the proponents present and discuss the results obtained from the study in relation to the stated  
objectives and research problems. The findings are analyzed, interpreted, and compared to the expected  
outcomes to provide a clearer understanding of AquaNova’s performance and effectiveness.  
Confusion Matrix of Object-Trash Detection:  
Table Iv Confusion Matrix of Object-Trash Detection  
Confusion Matrix of Object-Trash Detection  
Predicted Trash Type Actual Plastic Actual Metal Actual Non-Trash  
Plastic  
10  
0
3
6
1
1
1
8
Metal  
Non-Trash  
0
Table IV presents the Confusion Matrix of AquaNova’s Object-Trash Detection System, evaluating  
classification performance across plastic, metal, and non-trash categories using ten objects per type, for a total  
of thirty objects. All plastic objects were correctly identified, achieving ten out of ten. The non-trash category  
had eight correct classifications, while the metal category achieved six correct classifications, with three  
objects misidentified as plastic and one as non-trash. These results demonstrate AquaNova’s strong ability to  
detect plastic and non-trash items, while metal classification remains more challenging due to visual  
similarities with other materials.  
Table Ii Object Detection Accuracy Test  
Fig. 19 Accuracy Result (Confusion Matrix)  
In Fig. 19, 24 out of 30 total objects tested were correctly classified across the three categories. An 80%  
accuracy score confirms the system's strong capability to reliably identify floating waste and validates the  
effectiveness of the AI-based detection system in achieving its objective. While this demonstrates robust  
overall performance, misclassification issues, particularly within the Metal category (six out of ten correct),  
highlight the challenges in distinguishing certain waste types.  
Page 1998  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 20 Plastic Performance Result  
The results for the Plastic class in Figure 20, show an excellent recall of 1.00, indicating that the system  
successfully identified 100% of all actual plastic instances tested. However, the precision for the Plastic class  
was calculated to be 0.71 (ten out of fourteen), meaning that out of all instances predicted as plastic, only 71%  
were correctly classified. This lower precision is due to three actual metal objects and one actual non-trash  
object being mistakenly identified as plastic. The resulting F1-score of 0.83 represents a strong, balanced  
measure between precision and recall for the Plastic class.  
Fig. 21 Metal Performance Result  
Performance analysis for the Metal class in Fig. 21, which had seven actual instances, indicates a moderate  
level of detection. The calculated precision is 0.60 (6 out of 10), showing that only 60% of predicted metal  
objects were correct. This low precision results from false positives, including three plastic objects and one  
non-trash object incorrectly classified as metal. The recall, however, is higher at 0.86 (6 out of 10),  
demonstrating that 86% of actual metal instances were correctly identified. The F1-score of 0.71 reflects a  
reasonable balance between recall and precision while highlighting areas for improvement in metal detection.  
Page 1999  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Fig. 22 Non-Trash Performance Result  
In Figure 22, the analysis of the Non-Trash class highlights AquaNova's ability to protect freshwater species  
by correctly rejecting harmless objects. This class had nine actual instances in the test. The calculated precision  
for Non-Trash is 0.80 (8 out 10), meaning that 80% of the objects the model predicted as Non-Trash were  
correct. This strong precision indicates that the system generally avoids labeling waste as a non-target item.  
The recall for the class is 0.89 (8 out of 9), showing that the system correctly identified 89% of the actual Non-  
Trash instances, such as aquatic life, and allowed them to pass. The resulting F1-score is 0.84, the highest  
among the three classes, confirming the system’s reliability in successfully differentiating waste from non-  
waste items, which is a core objective of the AI-powered detection system.  
TABLE VCONFUSION MATRIX ANALYSIS  
Table V, shows the detailed performance metrics for the system. Plastic detection achieved high precision  
(1.00) but moderate recall (0.71), yielding an F1 Score of 0.83, indicating reliable identification with some  
missed instances. Metal detection displayed lower precision (0.60) but strong recall (0.86) and an F1 Score of  
0.71, reflecting difficulties in distinguishing Metal from other types. Non-Trash detection performed well, with  
precision of 0.80, recall of 0.89, and an F1 Score of 0.84. Macro-averaged metrics (Precision 0.80, Recall 0.82,  
F1 0.79) highlight AquaNova’s overall robustness in waste detection, with notable strengths in Plastic and  
Non-Trash classification and opportunities for improvement in Metal differentiation during collection  
operations.  
CONCLUSION  
The development of AquaNova, an autonomous solar-powered floating robot, successfully achieved the  
project objectives. The system integrates AI-based vision for real-time waste detection and classification, GPS  
waypoint navigation for precise autonomous movement, and a whale-inspired suction mechanism with  
Page 2000  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
segregated chambers for efficient waste collection and sorting. Automated return functionality ensures reliable  
and energy-efficient operation during prolonged deployment, demonstrating AquaNova’s capability to  
independently navigate, detect, and collect solid waste in aquatic environments.  
Testing showed the detection system effectively identifies and collects floating waste while avoiding  
freshwater species. Using a Raspberry Pi with OpenCV, YOLOv8, and a CNN model, AquaNova achieved  
80% overall detection accuracy. Plastic waste was detected with 100% precision, and non-trash items were  
correctly classified in 8 out of 10 instances. While metal detection had some errors, the results confirm reliable  
differentiation between waste and non-waste objects, supporting sustainable cleanup with minimal  
environmental impact.  
The GPS waypoint navigation system using the NEO-M8N module successfully enabled autonomous traversal  
and waste collection, fulfilling Objective 3. Reliability metrics showed a Mean Time Between Repairs of 25  
hours, Mean Time To Repair of 2 hours, and Mean Time To Failure of 23 hours, resulting in 92.6% system  
availability. Solar-powered support minimized downtime from manual battery charging. Overall, the integrated  
GPS and LoRa control system allowed AquaNova to efficiently remove floating waste, prevent blockages, and  
maintain continuous water flow in an eco-efficient manner.  
REFERENCES  
1. Asian Development Bank. (2024, February 28). Optimizing municipal waste collection in the Philippines  
through a smart app. Development Asia. https://development.asia/case-study/optimizing-municipal-waste-  
collection-philippines-through-smart-app  
2. Bupe Getrude Mutono Mwanza. (2025). Robotics in waste management: A smart technique to segregate,  
recycle, and recover electronic waste, 55–70. https://doi.org/10.1007/978-981-97-8673-2_4  
3. Carlo, I., Lecitona, E., Gamboa, M., Songco, W., & De Veas-Abuan, D. (n.d.). Implementation of  
filtering-waste robot for improved wastewater management control using Arduino. https:// www.  
dlsu.edu.ph/wp-content/uploads/pdf/conferences/research-congress-proceedings/2020/SEE-01.pdf  
4. Climate Change Commission. (2024). Ridge to reef: The fight against mismanaged waste. Climate.gov.ph.  
https://climate.gov.ph/news/923  
5. De Vera, Middle Tennessee State University. (2025, March 21). Mechatronics engineering B.S. degree.  
MTSU. https://www.mtsu.edu/program/mechatronics-engineering-b-s/  
6. DOST-MIRDC Galco Industrial Electronics. (n.d.). Relays. https://www.galco.com/ comp/prod/relay.htm?  
srsltid=AfmBOorzBsEUulH8ihFfz  
7. GitHub. (n.d.). About GitHub Desktop. GitHub Docs. https://docs.github.com/en/desktop/overview/about-  
github-desktop  
8. Gacu, J. G. (2023). Design of river floating trash traps using recycled plastic bottles and characterization  
of waste collected in Odiongan, Romblon Philippines. 10th International Conference on Recent  
Challenges in Engineering and Technology (ICRCET-22). https://www.researchgate.net/ publication/  
366987298_Design_of_River_Floating_Trash_Traps_Using_Recycled_Plastic_Bottles_and_Characterizat  
ion_of_Waste_Collected_in_Odiongan_Romblon_Philippines  
9. GeeksforGeeks. (2017, October 15). Understanding the confusion matrix in machine learning.  
GeeksforGeeks.https://www.geeksforgeeks.org/machine-learning/confusion-matrix-machine-learning/  
10. ISO. (2025). ISO/IEC 25051:2014. https://www.iso.org/standard/61579.html  
11. Gorecho, D. (2024, October 10). Turning the tide: Tackling plastic pollution through coastal  
cleanups,policy reform. Mindanao Gold Star Daily. https://mindanaogoldstardaily.com/archives/146703  
12. Gupta, S., Kruthik, H. M., Hegde, C., Agrawal, S., & Bhanu Prashanth, S. B. (2021). Gar-Bot: Garbage  
collecting and segregating robot. Journal of Physics: Conference Series, 1950(1), 012023.  
https://doi.org/10.1088/1742-6596/1950/1/012023  
13. Hariprasad, A., Shravya, & Sunil, T. (2021). Waste segregation robot - A Swachh Bharat initiation.  
IJISET - International Journal of Innovative Science, Engineering & Technology, 7. https:// ijiset.com/  
vol7/v7s12/IJISET_V7_I12_42.pdf  
14. Karthik, M., Manikanta, C., Vamsi, V., Reddy, K. S., & Bala, I. (2023). Autonomous river cleaning  
system using GPS technology. 2023 International Conference on Sustainable Computing and Data  
Communication Systems (ICSCDS), 1000–1003. https://doi.org/10.1109/icscds56580.2023.10104610  
Page 2001  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
15. Kestane, B. B., Guney, E., & Bayilmis, C. (2024, July 9). Real-time recyclable waste detection using  
YOLOv8 for reverse vending machines. Jurnal Ilmiah Teknik Elektro Komputer dan Informatika  
(JITEKI), 10(2), 345–358. https://doi.org/10.26555/jiteki.v10i2.29208  
16. Lasquety, M. K., Rosete, J. I. Y., Tan, K. A. D. M., Toco, M., & Blancaflor, E. (2022). iAMASS: An IoT  
design of a sewage control monitoring system. In 2022 4th International Conference on Management  
Science and Industrial Engineering (MSIE) (pp. 251–256). ACM. https://doi.org/ 10. 1145/353 5782 .353  
5848  
17. MacMillan, A., Preston, D., Wolfe, J., & Yu, S. (2020, May 19). Basic statistics—mean, median, average,  
standard deviation, z-scores, and p-value. Engineering LibreTexts. https://eng.libretexts.org/Bookshelves/  
Industrial_and_Systems_Engineering/Chemical_Process_Dynamics_and_Controls_(Woolf)/13%3A  
_Statistics_and_Probability_Background/13.01%3A_Basic_statistics-  
_mean_median_average_standard_deviation_z-scores_and_p-value  
18. Munir, S. (2022, October 18). Clearpath robotics launches outdoor autonomy software. Clearpath  
Robotics.https://clearpathrobotics.com/blog/2022/10/clearpathrobotics-launches-outdoor-autonomy-  
software/  
19. Prakash, C., Spoorthi, S., Pari, K., Jayanth, T., Naveen, V., Lavanya, C., & Bhalla, L. (2024). Design and  
prototype development of trash collector boat. MATEC Web of Conferences, 392, 01055. https:// doi.org/  
10.1051/matecconf/202439201055  
20. Pasig, Tullahan, other PH rivers are the top plastic-emitting rivers in the world – study. (2021, June 8).  
Climate.gov.ph. https://www.climate.gov.ph/news/508  
21. Ramirez, M. A., Rahman, S., Tan, S. W., Asyhari, A. T., Kurniawan, I. F., Mohammed, & Uddin, M.  
(2024). IoT-enabled intelligent garbage management system for smart city: A fairness perspective.  
22. Rekioua, D., Mokrani, Z., Kakouche, K., Rekioua, T., Oubelaid, A., Logerais, P., Ali, E., Bajaj, M.,  
Berhanu, M., & Ghoneim, S. (2023, December 9). Optimization and intelligent power management  
control for an autonomous hybrid wind turbine photovoltaic diesel generator with batteries. Nature. https://  
www.nature.com/articles/s41598-023-49067-4  
23. Samonte, M. J. C., Baloloy, S. H., & Datinguinoo, C. K. J. (2021). e-TapOn: Solar-powered smart bin  
with path-based robotic garbage collector. In 2021 IEEE 8th International Conference on Industrial  
Engineering and Applications (ICIEA) (pp. 181–186). IEEE. https://doi.org/ 10.1109/ ICIEA5 2957 .2021  
.9436763  
24. SEMCOR. (n.d.). How do conveyor belts work. https://www.semcor.net/blog/how-do-conveyor-belts-  
work  
25. Son, J., & Ahn, Y. (2024, December 16). AI-based plastic waste sorting method utilizing object detection  
models for enhanced classification. ScienceDirect. https://www.sciencedirect.com/ science/article/ abs/  
pii/S0956053X24006469  
26. Tee, M. L., & Cruz, D. E. (2022). A vehicle routing problem in plastic waste management considering the  
collection point location decisions. In 2022 IEEE International Conference on Industrial Engineering and  
Engineering Management (IEEM) (pp. 1–6). IEEE. https://doi.org/10.1109/IEEM55944.2022.9989979  
27. Unni, A., Gowri, S., Krishna N. A., & Livera, S. (2024). Issue 4 www.jetir.org (ISSN-23495162). Journal  
of Emerging Technologies and Innovative Research, 11. https://www.jetir.org/papers/JETIR2404F83.pdf  
28. Uzoma, D. (2022). Mechatronics-based waste collection and disposal system. International Journal of  
Innovative Science and Research Technology, 7(10). https://www.ijisrt.com/ assets/upload/files/ IJISRT  
22OCT046_(1).pdf  
29. World Bank. (2021, March 21). Philippines: Plastics circularity opportunities report. World Bank. https://  
www.worldbank.org/en/country/philippines/publication/market-study-for-philippines-plastics-circularity-  
opportunities-and-barriers-report-landing-page  
Page 2002  
www.rsisinternational.org