Trinetty Bot

The United Nations organisation’s 17 sustainability goals are meant to make the world more sustainable.The 6th target under the 11th goal is to reduce the adverse per capita environmental impact of cities, including by paying special attention to air quality and municipal and other waste management by 2030.

Trinetty Bot is a cleaning robot to improve waste management and reduce the labour of street sweepers.

The robot is called Trinetty Bot inspired by the Christian theological concept- Trinity which means “God is three in one” and is based on the French word ‘nettoyer’ which means to clean and 'Tri' which means ‘three’. The name of the robot tells the purpose of the robot in its full capacity.

To design the robot, I used 3D CAD modelling software Creo 3.0. The 3D-CAD modelling was done for the hardware body of the robot. For prototyping, I am to 3D print the robot in its entirety.The robot’s outer dimensions are as follows:

• Height: 300mm

• Length: 800mm

• Width: 400mm

Design for Floatation

To make the robot float, we need to consider both these aspects that make the body float on water. To reduce the average density of the robot, we need to make sure that the lower part of the robot is filled with air. Air has a density of about 1.2 g /litre, and water has a density of about 1 kg /litre. Air is therefore about 830 times less dense than water. This makes sure that our robot will stay afloat. But the problem is that we cannot pack the robot with compressed air to reduce the average density of the robot as that is an unpractical and unfeasible solution.

Instead, we use 2 techniques to increase the chance of floatation for the robot.

• Multihull design

• Make the body of polycarbonate

Multihull Design

Boats have a hull-shaped surface to make them float in water and increase the buoyancy force acting on the surface of the boat upward to oppose the gravitational force acting downward. The image below shows the shape of the hull used for boats that displace water as it propels forward.

Although the hull shape is good for floating, it tends to roll and is hard to incorporate into the robot design that needs to move on land with wheels. The table below shows the characteristics of various types of hull shapes.

From the table, we can conclude that the multihull design is best suited for our application as the robot needs added stability and manoeuvrability in the water. The higher contact surface area makes the design very stable for operation. Since the hulls are parallel and give a height to the body, it is best suited for placing the vacuum suction duct in the front.

The design incorporates 2 pontoons on either side to balance the robot in water. This enables the wheels to be unaltered as the parallel pontoons can enclose the wheels and make sure the robot works both on land and water.

Polycarbonate Body

Our design includes a multihull design for which we have that is made of polycarbonate material. This is because polycarbonate has higher buoyancy with a density of 0.92–0.98 g/cm3. Therefore, when polycarbonate is used in the hulls on both sides and the bottom surface of the entire length of the robot, the robot will stay stable and afloat while in water.

Propulsion Mechanism

To clean the water bodies, we have positioned the wheel between the robot body and the pontoons on either side. This gives space for the wheels to rotate freely on land while the pontoons enable floating in water. This gives the trust required for forward and reverses movement while the robot is in water.

The main propulsion for the robot is not the wheels though. The prime mover in the robot in water is the electric motor driven propeller drive. This is the best propulsion system for the robot as it does not hinder the other functionalities of the robot in both land and water.

Cleaning Mechanism

For the robot to achieve its purpose of cleaning the dirt and other wastes that are present in the road surfaces, homes and surface of the water, we use a vacuum pump that pumps the waste from the ground and also use the regulated storage inside the robot to store the collected waste and as a matter of the disposal of the waste. We can also open the top of the robot to take out the waste and dispose of it in the correct waste disposal area.

For cleaning the waste in water bodies, we use a rotating collector in the front of the robot. The collector will be able to collect the waste as the collector is in direct contact with the surface of the water as all the leaves or plastic will be able to enter the robot’s waste reservoir.

The waste detection was done by first creating a dataset of 1700 images from the college premises from the permission of the mentor and the principal. The dataset was collected by using an RC car with a camera mounted on it. The camera was able to capture the leaves present on the road. Videos of roads with leaves were recorded using the camera for dataset creation.

Dataset annotation

Once the videos were captured, the videos were converted into images by frame rate for using the images as dataset for the deep-learning model. To train the model, we first annotated all the images in the dataset by labelling each image for the class: leaf.

Annotation is the process of labelling each area of interest in every image in the dataset to train the model. The area of interest in each image associated with the label will be used in the model training process to learn the classification between the classes in the dataset.

This process is crucial and determines how precise and accurate the output of the model is. We used online tools for this purpose to make the process easy and enable remote access to the entire dataset for all members of the team.

Deep-Learning Model training

Once these images were annotated, we used the YOLOv7 object detection state-of-the-art algorithm to train our deep-Learning model for highest accuracy. We trained our model using the weights from the baseline of YOLOv7 model using transfer learning. This made the training speed to increase and accuracy to rise.

I trained the object detection model with the following specifications for highest precision and accurate results.·

• Batch size: 16·

• Epochs: 55·

• Weights: yolo_training.pt

The final output of the trained model results in the system being able to identify leaves on the road by itself without any human intervention. Such results will be used to trigger the cleaning mechanism if the waste is detected on the road.

Obstacle avoidance

To achieve obstacle avoidance, we use ultrasonic sensor to measure the distance of the obstacle in front and stop the robot accordingly. The obstacle might be a person or thing or anything that might hinder the pathway of the vehicle.

I designed the circuitry of the obstacle avoidance system using a Arduino uno board connected to ultrasonic sensors. I used the ultrasonic sensor to measure distance between the robot and the obstacle in front of the robot. The sensor we have designed for is ultrasonic sensor HC-SR04. It is a cheap and effective ultrasonic sensor.

The model also uses a micro servo motor in the design to rotate the wheels and is connected to the Arduino Uno to turn off if an obstacle is detected. The Arduino Uno is programmed in such a way that if the ultrasonic sensor senses an obstacle less than 50 cm the motors will turn off. The motors will run if the obstacle is beyond 50 cm.