Autonomous Floating Debris Collection

System for Water Surface Cleaning

K Subhashini K Shreya Shree A Abirami
Department of Electronics and Department of Electronics and Department of Electronics and
2024 International Conference on Communication, Computing and Internet of Things (IC3IoT) | 979-8-3503-5268-9/24/$31.00 ©2024 IEEE | DOI: 10.1109/IC3IoT60841.2024.10550252

Communication Engineering Communication Engineering Communication Engineering

(Associate Professor) (UG Student) (UG Student)
Sri Sai Ram Engineering College Sri Sai Ram Engineering College Sri Sai Ram Engineering College
Chennai , India Chennai , India Chennai , India
[email protected] [email protected] [email protected]

R D Kalaimathy R Deepika Dr. J Thamil Selvi

Department of Electronics and Department of Electronics and Department of Electronics and
Communication Engineering Communication Engineering Communication Engineering
(UG Student) (UG Student) (Professor)
Sri Sai Ram Engineering College Sri Sai Ram Engineering College Sri Sai Ram Engineering College
Chennai , India Chennai , India Chennai , India
[email protected] [email protected] [email protected]

Abstract— This paper presents the design and focus on efficient garbage detection using deep learning
implementation of an autonomous waste collection robot aimed technology.
at efficiently lifting and disposing of debris from water surfaces.
The robot utilizes a conveyor system, driven by a servo motor, Convolutional neural networks (CNN), in contrast to
to collect various waste materials, including water debris,
conventional detection techniques, get beyond constraints
garbage, and plastics. A water wheel, powered by a BLDC
motor, enables the robot's movement on water, enhancing its and allow improved object detection. Our contribution is a
cleaning capabilities. The system incorporates an accurate two-scale prognosis, specifically real-time detection and
computer vision system, leveraging an improved YOLOv5s reclustered anchor boxes to enhance accuracy, leveraging the
algorithm. The proposed solution employs a set of sensors, YOLOv3 algorithm, tailored to recognize plastic bottles,
including GPS, Compass, and a camera, for precise navigation bags, and styrofoam.
and waste identification. The paper outlines the hardware
components, including Raspberry Pi , Flight Controller , LiPo
Batteries, Transceivers, and modules, used in the system. II. LITERATURE SURVEY
Experimental results demonstrate the effectiveness of the robot
in autonomously cleaning water surfaces and safely collecting
An important development in the battle against
waste debris. This project represents a significant step toward contamination of the water's surface is the installation of
the development of a robotic platform for environmental rubbish collecting technologies. Recently, there has been a
conservation and river pollution management. growing awareness of the detrimental impact that trash has
on aquatic habitats, which has led to the development of
Keywords-Autonomous Waste Collection , Computer Vision innovative remedies. This section offers a thorough
, Sensors , Arduino . explanation of waste collecting methods and how they work
to effectively clean the water's surface. By reading through
I. INTRODUCTION the existing literature, we may gain more knowledge about
There is an estimate of 5.25 trillion particles of plastic in the the technical developments in the autonomous garbage
waters globally, out of which 269,000 tons remain suspended collecting industry. The discussion touches on recent
on the surface. In recognition of the critical need to clean up advancements, highlighting how these robots might
the environment, especially around urban water bodies, we revolutionize traditional methods of cleaning water and aid in
understand that routine trash removal has not been adequately the creation of a more sustainable plan for reducing water
overseen. Given the financial constraints encountered by pollution.
traditional river cleaning committees, we have put forward a
remote disposal truck. Three key components of a vehicle are By contrasting two creative initiatives in this area, different
needed to combat debris on banks and shores: Computer strategies for resolving waste management and water
Vision, Electric Device Control, and Mechanism on the pollution issues are shown. The "Blockchain-Based IoT-
Vehicle. Enabled E-Waste Tracking and Tracing System for Smart
Cities" project serves as a noteworthy example of the need of
The proposed research describes a semi-autonomous vessel utilizing state-of-the-art technology to address the complex
designed for gathering rubbish that floats. Innovative waste problem of trash management. The technology makes sure
collecting techniques are required due to the exponential that trash is continuously monitored and detected in
growth of plastic production, which exceeded 389.6 million metropolitan areas by using blockchain and IoT, which
tons in 2023 and Global plastic waste production is expected improves the effectiveness of waste collection and disposal
to triple by 2060, exceeding one billion metric tons, driven by procedures. Moreover, utilizing Python software to identify
increasing populations and economic expansion. The paper and separate items in real time is a progressive approach to
details the development of a garbage cleaner robot, with a trash management, demonstrating flexibility and adaptability
in managing various waste streams.The "Water Surface Solid

979-8-3503-5268-9/24/$31.00 ©2024 IEEE

Authorized licensed use limited to: Zhejiang University. Downloaded on July 21,2025 at 17:04:04 UTC from IEEE Xplore. Restrictions apply.
Waste Cleaning Robot For Ponds" project, in contrast, takes bodies of water as it moves. The waste debris in the water will
a more targeted approach, focusing on particular locations be uplifted as the machine is submerged and will travel
like ponds for garbage collecting. The project highlights upward. The waste debris will fall into the tray as soon as it
efficiency and simplicity in garbage removal from water reaches the uppermost position. The base structure holds a
surfaces by creating an autonomous trash collector robot shaft that the water wheel is bolted to. A water wheel's
outfitted with a radio frequency module and conveyor belt function is to propel the apparatus forward or backward
technology. This method not only takes care of the urgent across water. As a result, the water's surfaces will be cleaned
need to clean the pond, but it also offers a scalable solution and waste debris will be effectively eliminated from the
that can be used for other bodies of water facing comparable water. The water wheel is rotated by a motor. The conveyor
problems with garbage accumulation. for collecting waste is driven by a servo motor. Ultimately,
the tray that is positioned inside the boat gathers the floating
It is possible to get important insights for the creation of an trash.
autonomous floating debris collecting system by contrasting
these two literature reviews. The first project's use of cutting-
edge technologies like blockchain and IoT can improve the
system's capacity for thorough waste management and
tracking. Conveyor belt systems and radio frequency
modules, two useful design components from the second
project, may be combined to maximize the system's
effectiveness and versatility in a range of water surface
cleaning situations. This combination of cutting-edge
technologies and workable design solutions has the potential
to significantly reduce water pollution and encourage
environmental preservation.
III. EXISTING SYSTEM
The current approach to river cleaning can be accomplished
through the utilization of river skimmers.River Skimmer,
which floats on the water's surface, sucks river water in,
screens out debris, and discharges clean water, is a method to
use the system for cleaning the rivers. After the plastic
storage container is filled, it is removed for recycling. A
disadvantage is that they need to be expertly observed at all
instances and are susceptible to becoming blocked by ice and
debris.Trash rakes are equipment that are used to collect and
lift floating rubbish from the water. They're equipped with
multiple rows of spaced bars or teeth. The trash can be
extracted by hand from the rake once it was properly
gathered.Personnel manually eliminates floating garbage
from the water's surface through hand nets or scoops.
Sometimes barricades or floating booms are put in position
precisely to confine waste to make it easier for workers to
retrieve it.

However, former systems relied primarily on standard Fig 1 - Flowchart of the Proposed System
sensors to determine rubbish; they were manually controlled
and lacking complete automation. Furthermore, especially in An autonomous waste-cleaning robot may be an effective
more expansive water garbage regions where real-time instrument for successfully mitigating river pollution. For a
effectiveness is crucial, many stationary systems do not cleaning robot, identifying and tracking trash are the most
succeed to meet the basic requirements for an effective river important aspects of the machine's visual system. A key
cleaning system. component of a robotic platform is the development of an
accurate computer vision system. The purpose of this project
IV. PROPOSED SYSTEM is to acquire visual data regarding the floating trash in the
river. The complexity of the environment makes it difficult
Our system's primary purpose is to eliminate waste
for traditional image processing methods to meet the
particles from the water's surface and deposit them in the tray.
requirements of real-time monitoring of floating debris in the
A gear motor, a Raspberry Pi, a motor driver, a conveyor belt,
waterway.
two 12V and three 3.7V batteries, jumper wires and PVC
pipes make up the initial concept. The prototype is built in To deal with these challenges, an enhanced YOLOv5s
such a way that the conveyor belt rotates as soon as the boat algorithm is proposed, which adds a feature map attention
moves. The conveyor arrangement is mounted on the motor (FMA) layer at the end of the backbone. The network,
shaft. The conveyor rotated as a result of the motor's rotation. training, and detection operations make up this method.The
The conveyor gathers plastics, rubbish, and debris from model's weight parameters are updated on a regular basis, and

Authorized licensed use limited to: Zhejiang University. Downloaded on July 21,2025 at 17:04:04 UTC from IEEE Xplore. Restrictions apply.
the model's weight value remains constant throughout 2. Telemetry Module:The antenna that's mounted on the
detection. The final result is chosen and indicated on the telemetry module is 433MHz in frequency. The ground
image after the predicted value has been filtered based on the station, a laptop furnished with a 433MHz USB dongle,
threshold value. communicates to the flight controller using the transmitter as
an input/output device. The operator can obtain data from
A. Mechanical Design :
both internal and outside sensors and remotely modify
The prototype functions by having the conveyor belt start to parameters. The boat can also be directly controlled using this
rotate when the boat moves forward. It has a setup where a connection.
conveyor is mounted on the motor shaft. Waste debris in the
water is raised and moves upward when the conveyor spins
due to the motor's rotation. When it reaches the top, the waste
debris is placed within the tray. The machine is propelled
forward or backward on water by a water wheel that is fixed
to a shaft on the base frame. In the end, the floating waste is
collected by the boat's tray.
Fig 4 - Telemetry Module

3.RC Receiver and Remote: The receiver has a connection to

the remote control. Through its six channel capacity and two-
way communication feature, the RC boat is capable of getting
signals via the receiver. Using the I-bus protocol, the output
signals from the receiver are sent to the controller across a
single wire

Fig 2 - Hardware Implementation

Specifications Dimensions

Fig 5 - RC Remote Control

Length × Width × Height 75 cm×48 cm×24cm
Weight 6 kg 4.Video Transmitter (VTX ): The flight controller accepts the
Trash payload 1 kg video signal via the camera and adds an on-screen display
Maximum Speed 1.3 m/s (OSD) to the video stream to show data like battery voltage.
After that, it gets transmitted to the video transmitter, or VTX,
Table 1 - Proposed Design Specifications which delivers it to a 5.8GHz receiver on the other end. While
B. Autonomous and Electronics Component Setup : not absolutely necessary, it's intriguing to be able to see what
the boat is photographing.
a) Main Hardware Components:
1.Flight Controller: The flight controller manages the
ArduPilot Rover Firmware. This software can connect to
sensors and input as well as output devices, monitor battery
voltage as well as current, function motor controllers via
PWM outputs, and integrate a gyroscope, among other
capabilities. The flight controller acts as the system's "brain" Fig 6 -Video Transmitter
in concept.

5. GPS and Compass Module: A gadget which combines GPS

and compass the ability into a single module is the GPS and
Compass module M8N. It makes use of an HMC5883L
compass module and a UBlox M8N GPS module. This
component is helpful for a wide range of applications,
involving drones, robots, and other autonomous systems, as
it can give precise positioning and heading data to aid
Fig 3 - Flight Controller
navigation. In addition, the M8N features an integrated
compass that may be used to offer heading information. This
feature can be especially helpful when the device is stationary

Authorized licensed use limited to: Zhejiang University. Downloaded on July 21,2025 at 17:04:04 UTC from IEEE Xplore. Restrictions apply.
or moving slowly. recognising trash and debris in multiple environmental
contexts.
The YOLOv5 model is trained using the TACO dataset,
which gives the algorithm the capacity to reliably identify and
categorize various kinds of floating debris. The dataset
ensures that the trash collector can efficiently identify and
collect waste materials across a range of scenarios frequently
encountered in river and water body cleanup initiatives,
hence contributing to the model's robustness and
generalizability.
Fig 7 - GPS and Compass Module 2. Path Planning for Coverage on Water Surfaces
b) Software Modules:
The Flight Controller is initially loaded with the latest
The software for this river trash treatment project consists of Ardurover firmware (version 4.0.0) for seamless integration.
an extensive microcontroller programming framework
The FRAME_CLASS is configured as 2, specifically
written in C or C++ that synchronizes the operation of several
system components. Sophisticated software in the design of designed for boats, and the requisite parameters are
the remote control interface makes it easier to understand how appropriately configured. The ground station, which can be
to interpret commands from the RC remote to control the any PC, hosts the Mission Planner software. Way-points are
trash collector's motions and functions. In computer vision strategically marked manually to define the cleaning area
projects, visual data from cameras or sensors is processed path of the boat.
using an algorithm that makes it easier to locate and identify
floating debris. Using GPS or similar location technology, the The cleaning area may contain known obstacle zones, each
navigation system's software carefully plans routes, avoids represented by a set of bounding points represented as O =
obstructions, and makes sure the trash collector stays in {B1,B2,B3…..Bn }, where 'n' denotes the number of points. The
designated collection zones. Furthermore, power path planning method requires inputs of the boundary point
management software systematically optimizes energy use, polygon of the cleaning area, along with the set O
especially in systems.The software architecture of the river representing the boundary points of known obstacles. The
waste cleaning project is intricately designed to incorporate method outputs the planned optimal path L, consisting of a
advanced features, including the integration of the YOLOv5 series of path points' coordinates.
(You Only Look Once version 5) algorithm for efficient
debris detection. From the Coverage Path Planning (CPP) Algorithm , to
1. YOLOv5 simultaneously achieve shorter path and fewer moves we can
For accurate identification and localization of floating debris, deduce the mapped path planning efficiency as
the YOLOv5 method is essential to real-time object ‘η’,representing the total time taken to complete the task
detection. By including a feature map attention (FMA) layer, according the planning path P ,
this system has been improved to solve real-time monitoring
issues in complicated situations. This sophisticated algorithm η(P) = D / s + i * Tavg , (1)
drives the navigation system, which uses GPS or other
comparable positioning technologies to carefully plan routes, where D denotes the total path length P , s denotes the
avoid obstructions, and make sure the trash collector stays in average velocity of the robot, Tavg denotes the average
designated collecting zones. turning time, and i denotes the number of planned turns.
The rationale behind selecting YOLOv5 was its ability to
strike a compromise between efficiency and performance,
which makes it ideal for real-time applications such as
autonomous trash collecting on water surfaces. Compared to
YOLOv8, its simplified design enables shorter inference
times, which is essential for our robot's responsiveness in
quickly and effectively recognizing and gathering
garbage.Furthermore, the adaptability and resilience of
YOLOv5 in managing a range of object detection tasks, such
as garbage recognition, complement the project's diverse
scope. Its real-time capacity to precisely identify and
categorize various debris kinds adds to our autonomous
cleaning system's overall efficacy.
The TACO (Trash Annotations in Context) dataset is utilized
by the river waste cleanup project in addition to the YOLOv5
algorithm. This dataset is essential for training and testing the Fig 8 -Schematic diagram of waypoints set around cleaning
model's object detection skills. The TACO dataset offers a area in Ardupilot
large collection of annotated photos that include a variety of
waste material types, lighting situations, and complicated The designated home location, where the boat is armed,
backgrounds. It is specifically selected to solve the issues of serves as the reference point for the Return-To-Location

Authorized licensed use limited to: Zhejiang University. Downloaded on July 21,2025 at 17:04:04 UTC from IEEE Xplore. Restrictions apply.
(RTL) command. Consequently, the boat consistently returns
to this location upon receiving the RTL command. The
autonomous navigation begins from way-point one,
progressing through subsequent way-points to cover the
predefined path. Surveillance capabilities are enhanced
through the integration of an HD camera.

V. DISCUSSION OF RESULTS
Fig 10- Losses vs Epoch Graph
Table 2 - System Characteristics of the Trash Collector
VI. CONCLUSION
The YOLOv5 model performs well with low loss values
Functional Characteristics Hardware Used
across all components, demonstrating effectiveness in real-
time monitoring of floating objects in waterways.To improve
Processor Raspberry Pi 4 Model B +
robustness in aquatic environments, we need a broader
Flight Controller
dataset, particularly focusing on enhancing detection of
blurred and dense objects, and expanding support for
Sensors RGB HD Camera , GPS
additional floating debris categories.The proposed system
module
was able to lift solid debris using a conveyor belt, and the bot
Power Supply 24V 140 AH LiPo Batteries was lightweight and portable.This project significantly
, Solar Panels decreased solid waste and enhanced workers' lives by
offering comfort and ease through smart monitoring,
Communication LTE USB Modem reducing manual cleaning efforts.

Control Mode Automatic / Manual / VII. REFERENCES

Return to Launch (RTL)
[1] Statista. (2023b, November 20). Forecast global plastic
Running Time 8 hours waste generation 2019-2060.
[2] Patil, C., Tanpure, S., Lohiya, A., Pawar, S., & Mohite,
P. (2020, November). Autonomous Amphibious Vehicle
Table 3 - Evaluation metrics of the implemented model for Monitoring and Collecting Marine Debris. In 2020
5th International Conference on Robotics and
Precision 74% Automation Engineering (ICRAE) (pp. 163-168). IEEE
[3] Satheesh, J., Nair, A. P., Devipriya, M., Chithra, A.,
Accuracy 76% Mahesh, G., & Jayasree, P. R. (2020, June). Wireless
Communication based Water Surface Cleaning Boat. In
F1 Score 0.73 2020 4th International Conference on Trends in
Electronics and Informatics (ICOEI)(48184) (pp. 716-
Recall 0.71 720). IEEE..
[4] Steccanella, L., Bloisi, D. D., Castellini, A., & Farinelli,
mAP 0.64
A. (2020). Waterline and obstacle detection in images
from low-cost autonomous boats for environmental
monitoring. Robotics and Autonomous Systems, 124,
The YOLO version of the trained model attained an accuracy
103346.
rate of approximately 76%. It can successfully identify waste
with a confidence level of 0.65. Consequently, the model is [5] Tan, G., Zhuang, J., Zou, J., & Wan, L. (2022). Multi-
suitable for deployment in real-time classification of floating type task allocation for multiple heterogeneous
solid waste in rivers. unmanned surface vehicles (USVs) based on the self-
organizing map. Applied Ocean Research, 126, 103262.
[6] Jogi, N. G., Dambhare, A., Golekar, K., Giri, A., & Take,
S. (2016). Efficient lake garbage collector by using pedal
operated boat. International Journal of Modern Trends
in Engineering and Research, 2(4), 327-340.
[7] Kanade, P., Alva, P., Prasad, J. P., & Kanade, S. (2021,
April). Smart garbage monitoring system using Internet
of Things (IoT). In 2021 5th International Conference on
Computing Methodologies and Communication
(ICCMC) (pp. 330-335). IEEE.
Fig 9- Model’s Predicted Output [8] Renjith, R., & Reshma, R. (2019, December). Design
and implementation of automatic garbage collecting
robot using background subtraction based computer

Authorized licensed use limited to: Zhejiang University. Downloaded on July 21,2025 at 17:04:04 UTC from IEEE Xplore. Restrictions apply.
 vision algorithm. In 2019 9th International Symposium
on Embedded Computing and System Design (ISED) (pp.
1-5). IEEE.
[9] Yuan, B. H., Gao, X. H., Zhang, B., & Lin, H. Y. (2017).
The Design and Realization of Small-sized Water-
surface Garbage Cleaning Boat. Journal of Long dong
University, 28(1), 24-29.
[10] Prasad, P. R., Avinash, J. L., Kumar, G. A., Poornima,
G. R., Kumar, S. S., & Kumar, K. S. (2020, November).
Iot based smart water quality monitoring and flow
control system. In 2020 International Conference on
Recent Trends on Electronics, Information,
Communication & Technology (RTEICT) (pp. 358-362).
IEEE.
[11] Dhande, K. V., Ballade, A., Garde, V. S., Lahane, A. S.,
& Boob, P. V. (2017). Design and fabrication of river
cleaning system. International Journal of Modern trends
in Engineering and Research, 4(2), 8-18.

Authorized licensed use limited to: Zhejiang University. Downloaded on July 21,2025 at 17:04:04 UTC from IEEE Xplore. Restrictions apply.