Marino Bot

https://youtu.be/JktqammhfJI

https://docs.google.com/presentation/d/17H9t6d2vN4RJyB94MWGeNt7rUc9SnkFU/edit?usp=drivesdk&ouid=116717667533330298459&rtpof=true&sd=true

Project Description

Problem Statement

According to a study done by the waste management center in Abu Dhabi, the amount of waste generated in UAE is around 4.892 million tons. In daily basis, Abu Dhabi produces an average of 6935 tons of waste everyday, and 4118 tons in Al Ain. It was estimated by Ellen MacArthur that the amount of plastic waste in the sea will increase to a level where it will be more than the fish by 2050. Even the smallest creatures of the oceans, which is the sea horse, could not be able to evade the plastic waste. This problem is affecting marine lives and humans. With the technological advancements in the field of robotics, the world is currently moving towards the utilization of robots in various industrial fields. Robotics were introduced as a solution for waste collecting; however, most of them are remote controlled and unsustainable. As a result, we intend to design an autonomous, and sustainable sea cleaning Marino Bot, which will be able to clean up the water surfaces safely and continuously.

Motivation

The coastal waters and oceans are relapsing due to several reasons, such as pollution from the ships, increasing coastal development and land-based sources of pollution. Therefore, as a team of electrical and computer engineers, we were motivated to take a role in solving such a global issue by developing a completely autonomous and sustainable sea cleaning robot. Since technology nowadays is moving towards creating a world where intelligence is the dominant factor, we wanted to be a part of this world by proposing the Marino Bot, which depends on multiple machine learning algorithms, such as, object detection and object tracking. Additionally, governments are now trying to move their main energy consumption resources from natural gases to renewable resources, such as wind and solar energy. Therefore, we were driven by these worldwide gradual changes, and decided to include sustainability as an important factor our design by making our robot runs only on solar energy.

System Overview

Energy Management Systems and Methods

As the worlds energy consumption rate is increasing every year, energy management methods and procedures are being set and followed to limit the individual power consumption. In general, the main renewable source of energy in the UAE is solar energy, hence solar systems are used more frequently in the country. There are two main types of solar systems, regarding their connection, which are the off-grid and grid-connected systems. In off-grid systems, the battery is considered as the power bank that supplies all connected appliances with the needed power whereas on- grid systems do not require a battery to store the power generated from solar, rather it directly feeds the loads.

Object Detection (Regional-CNN)

There are a wide variety of methods that serves the purpose of object detection; However, every system has its own constrains that guide the developer to choose the optimal method. Convolutional Neural Networks (CNN) has high performance/accuracy with applications related to image classification; however, CNN is too slow and computationally expensive as it depends on sliding a window (convolving every region) through the whole image for detection. A better approach was proposed by a team in UC Berkeley, led by Professor Jitendra Malik which is Region Based Convolutional Neural Network (RCNN) where they used an algorithm called selective search that decreases the number of bounded boxes given to the classifier to 2000 only. Selective search depends on the idea of providing region proposals that might contain objects. A team at Microsoft Research developed an approach called Faster RCNN which is faster than RCNN by 250 times. The first part of the faster RCNN is the base network which is a deep fully convolutional network such as VGG or ResNet pretrained on ImageNet. This layer has a combination of convolutional Layers, Pooling Layers, and Fully Connected Layers. Usually, the first layer learns the edges then the second one learns patterns of the edges to identify more complex shapes and so on. At the end, we get a convolutional feature map that has more depth than the original data based on the number of filters it learns. However, its height and width decrease as a result of the pooling.

Object Tracking

Object tracking is used to follow the movement of the objects that the user is interested in, and to have the system predicts the location of the object in the next time slot. Kalman filter was discovered by R. E. Kalman in 1960 which focuses on the problem of extracting useful data from noisy measurement variables. Kalman filter is a powerful algorithm for filtering (smoothing) the input signal and predicting the next state. The input data which in our case are images will keep flowing to the kalman filter model. At the same time, the Kalman filter process a series of state estimations and measurement update steps to keep track of the future state of the object .

Impact Statement

Marino Bot aims to help revive the aqua environment by freeing it from all types of floating waste on its surfaces. It will help people working in the water surfaces cleaning field by significantly reducing their long working hours. Due to its autonomous nature, the robot is best fit for its independent cleaning. The business in sea cleaning will be affected, and companies might replace their methods in sea cleaning with our design. Additionally, our design of sea cleaning robot is sustainable, which is mean it does not depend at all on fossil fuels. Thus, the emissions of the greenhouse gasses will be decreased in the sector of sea cleaning. It is completely powered by solar energy, which is a clean, reliable, and sustainable source of energy and it’s considered to be a free energy source, so it decreases the energy consumption level of the economy. In addition, Our project misses any equipment or material that might pollute the air. Consequently, the robot is designed to take waste collection trips of the lowest energy cost, hence energy management and reduced consumption is maintained through the complete lifetime of the robot.

 Design

The robot is designed to harvest solar energy, locate wastes in water surfaces, and plan the shortest path to collect them. The design is implemented in two models to test all parameters possible on the prototype model then safely apply them on the real-life model. Our design faces some complications when it exposed to the real-life environment, which is sea. Sea waves are high in amplitude and fast, this will affect the energy consumption of the robot and that’s why we use the smart energy management using Travelling salesman technique to avoid those paths. In addition, we need our circuit to be fully waterproof as is might sinks with the present of waves.

-       System Overview

The system will start by capturing aerial images for the area of interest using a drone. Then it will send these images to the computer for processing. The system detects the objects which are the robot, waste and obstacles using Faster-RCNN. We choose Faster- RCNN because it has low computational cost compared to the other approaches as well as it has a much higher speed. After the detection and classification, we use object tracking to track the movement of the robot and the waste. The approach we use is Kalman filter object tracking which fits the requirements of our model. It can detect multiple objects, estimate the next state even when occlusion occur and predict the movement of a linear system. Energy management is one factor that decides whether the robot should start the waste collection trip or not. If the solar battery has the trips energy demand without exceeding its limit, which is 80% of its stored energy, then the robot is allowed to start the trip; otherwise, it should wait until it collects the energy demand first. Energy cost of a specific robot-waste location scenario is first calculated between all points through a generated formula by regression algorithm provided on MatLab. The energy model is generated based on previous data of robot trips and the energy expenditure in each one. Using energy model, cost between the robot and all way points and between each two points in a path will be identified, thus now we can use Dijkstra's algorithm to find the best path that the robot can take. This algorithm defines a shortest tree path set, which finds short paths from a basic initial vertex to different vertices and ending with a last stop vertex point, thus each path in this tree has different cost. Moreover, the system implements three PID controllers to assist the robot in navigating its way through the water. The first PID is for bearing control, second for thrust control, and a PID for maintaining a straight-line movement.

Conclusion and Future Work

In conclusion, the Marino Bot is able to clean water surfaces from all types of distinguishable floating wastes while following a system that applies a chain of algorithms, their main purpose is to direct the robot to take the path of least energy consumption while avoiding obstacles, collecting the largest number of wastes possible, and ending the trip at a sunny spot at daytime to recharge the battery through the implemented PV stand-alone system. The design of the robot is first done as a prototype where all algorithms are tested on, while the real-life-scaled robot design is identical to the prototype; however, its size is scaled up, thus the components used such as motors and controller boards are substituted with ones of higher and more sufficient ratings. The tests done on the prototype were dedicated to generalize an energy model, object detection and object tracking algorithms and models, path planning algorithm, and solar system general size and ratings, all to have an easier and secure implementation on the real-life model.

Recommendations

Our recommendation for this robot based on the results. We can enhance the accuracy and efficiency of the systems by the following recommendations:

The design can be more attractive for marketing.

• The design should be more robust, to handle the waves and wind test more accurately.

 • The design needs to be scaled up in size to fit the solar cells.

 • The software systems can always be updated, as the programming world keep updating and the AI in lasting growth.

• The Electric circuit equipment should be high quality, to reduce the maintenance and save money. The Electric circuit also should be protected from the water 100%.

 • At the bad weather, strong winds and high wave the Robot should not work to be safe. 

We utilized National Geographic's Debris Tracker Application's open citizen public database for Machine learning model training and NASA satellite data for obtaining accurate locations where high concentration of marine debris is found . We also utilised NDAA's facts on marine debris, US Department Interiors's Impacts article for gaining inspiration and as part of research work to do the project.

Working on this challenge encouraged us to push our boundaries and collaborate to come up with an optimum technical solution. Since we only had 48 hours, it was a nerve-racking journey, but we managed to pull through it successfully. Our group comprises of Electrical and Computer Engineers, hence we ensured that our project includes Artificial intelligence along with the sustainability concept. On referring to the examples provided on the NASA SpaceApps Challenge website, we were sheer motivated to find a solution to the leveraging marine debris problem. Our team collaboration skills helped our complete the project on time. We would like to thank Dr. Ghazal and Ms. Maha, our mentors for helping us with this project, our parents for supporting us. We would also like to acknowledge NASA Space Apps Challenge for giving us a delightful opportunity that will allow us to boost our career and will help us apply all this information into our future projects. Overall, it was a wonderful lifetime experience.

Data:

Public Database of National Geographic's Marine Debris Tracker App

Tools:

Matlab- Machine Learning (Object detection and Object Tracking)

Easy EDA - PCB Design

Multi Sim Software: Hardware Schematic

Draw.io- Flow Charts

MS PowerPoint- Presentations and Video

#Machinelearning #fastRCNN #objectdetection #objecttracking #pidcontroller #marinedebris #robot #artificialintelligence

This project has been submitted for consideration during the Judging process.