To successfully operate SWEEP, we extensively used ROS for algorithm
designing and testing, AI/ML for waste segregation and traversal,
Various sensors involved in sensing wastes, obstacles etc.
With help of these we developed an autonomous Algorithm for ultimate
movement which forms the core of the robot.
ROS is a pseudo operating system mainly used in Linux OS. It provides
various tools like the RViz, Gazebo etc. ROS greatly enhances our
capability for simulation and testing of our models in real world like
scenarios including obstacles, different paths etc.
We have done the main algorithm and design testing using ROS.
It is a 3-D Simulation tool provided by ROS. It provides a real world
like scenario by incorporating a programmed physics engine. We used
Gazebo to launch our designed model and various individual parts into
its pseduo real world and testing such as autonomous navigation
algorithm check was done by placing obstacles along the defined path.
It helps to do these testing without having to actually build any
parts.
Sensors help to perform action based on the input feedback data in
real time.
We have used HCSR04 ultrasonic sensors for purpose of obstacle
avoidance.
To achieve waste segregation we have used capacitive proximity sensors
which helps in sorting waste based on dielectric properties of various
materials.
QMC5883L sensor integrated with GPS and RPI-4 was used for orientation
purpose based on user input
The Bot incoprorates sophisticated integration of various algorithms
to achieve its desired goals.
Area coverage algorithm is a state of the art algorithm to efficiently
cover the user defined area guided by GPS.
Obstacle avoidance algorithm helps in changing the trajectory based on
the obstacles encountered in the vicinity.
Orientation algorithm enables the bot to orient itself to a desired
direction and alter its course of trajectory.
We have mainly explored the world of Deep Learning, which was used for
the segregation part.
A highly efficient and accurate model was built which was tested on
numerous data sets of different categories of wastes.
The input for the model was taken through a camera and a smart control
system was used which determined the type of waste using the output of
the Deep Learning Model and segregated it accordingly.
