MARINO BOT

Supervisors: Dr. Mohammed Ghazal & Eng. Maha Yaghi

�Members:

Farah Shaik Nawal Mehmood

Saba Parvez Zayna Wasma

Ishita Maharana

Challenge: Leveraging AI/ML for Plastic Marine Debris

​

Public Database

Of Debris Tracker App

Faster- RCNN

for object detection

​

Object Tracking Using �Kalman Filter

SYSTEM �OVERVIEW

Matlab’s Linear Regression to produce an energy model

Energy Management & �Path Planning

Select shortest

path using

Dijkstra Algorithm

Marino Bot begins

movement

Marino Bot Design Layout

The 3D Model of the robot’s chassis.

Dimensions: 55.3mm x 28.0mm x 12.9mm

The Printed Circuit Board (PCB) wired connections and the completed design to limit the number of wires used for the robot construction.

The hardware schematic of the circuit comprises of WiFi Feather M0- micro controller, L298N H-bridge module, two micro gear-motors (each of 6V and 0.33A rating), and a 9 Degree of freedom IMU sensor for PID control.

AI/ML Implementation

Fast RCNN machine learning Architecture is used for object

detection.

The PID controller is deployed to regulate speed of movement and ensure obstacle avoidance such as pole or a rock.

PID Controller Algorithm Flow Chart

Fast RCNN �Flow Chart

Faster RCNN-�Transfer Learning

A

B

Google Net – 5 Epochs

The detector trains for 5 epochs & can detect objects floating on the water surface to limited precision.

Google Net- 10 Epochs

The detector trains for 10 epochs to detect the floating objects with higher precision and accuracy.

It saves time and resources since the network need not be implemented from scratch but can be tuned from previous models.

Expected result after training indicates that certain objects are failed to be detected and labeled.

Expected results after training indicate that all objects are successfully detected and labeled.

Expected Demo Of The Implemented System

MARINO BOT

Challenge: Leveraging AI/ML for Plastic Marine Debris

​