mdp-blog2019

[mdp-blog2019]

Prototyping stages of the UAV

Upon the creation date of this report, our group has already finished the first plywood prototyping and an acrylic prototype with the new design of the top cover. The purpose of these two prototypes is to check the dimension of all screw holes between top and bottom planes.

However, we found one issue with the frame is that the arm which supports the motor will vibrate when the motor spin, this will greatly decrease the durability and stability of the arm. Therefore, we decided to attach a QAV500 G10 arm on top of the original arm to enhance the structure.

The image below shows the first prototype. The propellers are fully protected by the frames and the arms are enhanced by G10 carbon fiber arms which we adjust the dimension for the frame to perfectly fit the arms.

The ESCs will be placed on top of the arms so that the propeller can cool down the ESC during the flight.

The flight controllers will be placed in the center of the bottom frame for optimum balance and performance.

The next step is to cut the bottom plate with carbon fiber to have better strength.

First wooden Prototype with enhanced arms

Acrylic Prototype with all 4 arms, motors and propellers installed

[mdp-blog2019]

SAFMC UAV

UAV Block diagram

UAV DESIGN

For the maximum space for tolerance, we decide to design a drone with 360° full coverage prop-guard to allow the drone collide with the wall without damage.

The following figures are the Computer Aid Drawing of the drone frame.

UAV electronics

Motors:

The motor used are 4x AIR 2213 920kv motors

Motor Specifications

KV: 920

Configuration: 9N12P

Stator Diameter: 22mm

Stator Length: 13mm

Shaft Diameter: 4mm

Motor Dimensions(Dia.*Len): Φ27.5×30mm

Weight (g):  54g

ldle current(10)@10v(A): 0.5A

No.of Cells(Lipo): 3-4S

Max Continuous current (A) 180S: 18A

Max Continuous Power (W) 180S: 230W

Max. efficiency current: (3-10A): >83%

Internal resistance: 132mΩ

T9545 propellers were used,Plastic Propeller T9545A set (CW & CCW) with self-tightening function.

AIR 20A ESCs were installed which support a signal frequency of up to 621 Hz with special core program for multi-rotor controllers.

Flight Controller Specifications: (Previously)

F405 Processor

MPU6000 Gyro/ACC Sensor

BEC Max current 5V 1.5A

Built in OSD with max Current of 145A– configurable via Betaflight GUI

Power distribution board with six-layer PCB

Current Sensor

[mdp-blog2019]

SAFMC 2019 CAT D1

The MDP group will be participating together in the competition in late March to early April

Target & Requirements

Design and build a semi autonomously small air platform to perform a multitude of tasks in a small indoor space.

Category D1 requires its participating teams to race against each other through a series of obstacles and plays the game Tic-Tac-Toe by picking up and releasing payloads with a semi-autonomous platform.

Competition Plan

Referring to the information from the competition booklet, the distance between two walls is 2 meters and the circle at starting point has a diameter of 1.5 meter. The circuit is long with 3 ninety-degree corner so high tolerance allowance must be achieved.

The final goal is to play tic-tac-toe against another team and each team has 3 bean bags as payload. Both picking-up and dropping cannot have human interaction. Therefore,  a grabbing claw is required and it need to have the same height as the landing gear so it can touch the ground to has maximum efficiency of grabbing.

Due to the limitation of one VTX, we are going to connect 2 cameras to the VTX. One is at front of the bottom frame, pointing straight forward,  and one is at the back, monitoring the payload.

[mdp-blog2019]

Computer Assisted Detection System (Existing Computer Vision and image recognition methods)

OpenCV (Computer Vision Library)

OpenCV (Open source computer vision) is a library of programming functions mainly aimed at real-time computer vision.The library is cross-platform and free for use under the open-source BSD license.This library is used by the next few detection system

Using OpenCV 4.0.0 in conjunction with Visual Studio 2017 ,we are able to put video feed of the FPV camera into a C++ programme that will run it through pre-trained image classifier (haar cascade files) that detects human body.This is supposed to assist in a search and rescue.

OpenCV's application areas include:

2D and 3D feature toolkits

Egomotion estimation

Facial recognition system

Gesture recognition

Human–computer interaction (HCI)

Mobile robotics

Motion understanding

Object identification

Segmentation and recognition

Stereopsis stereo vision: depth perception from 2 cameras

Structure from motion (SFM)

Motion tracking

Augmented reality

To support some of the above areas, OpenCV includes a statistical machine learning library that contains:

Boosting

Decision tree learning

Gradient boosting trees

Expectation-maximization algorithm

k-nearest neighbor algorithm

Naive Bayes classifier

Artificial neural networks

Random forest

Support vector machine (SVM)

Deep neural networks (DNN) 

TensorFlow (Machine Learning Library )

TensorFlow is an open source software library for high performance numerical computation. Its flexible architecture allows easy deployment of computation across a variety of platforms (CPUs, GPUs, TPUs), and from desktops to clusters of servers to mobile and edge devices. Originally developed by researchers and engineers from the Google Brain team within Google’s AI organization, it comes with strong support for machine learning and deep learning and the flexible numerical computation core is used across many other scientific domains.

TensorFlow is used with OpenCV to train the computer to learn and recognise a specific thing which the user desire

YOLO,You Only Look Once (Image recognition method)

Currently the fastest Open source image recognition system ,YOLO which is able to use graphic processing unit to process its image unlike the other current conventional methods where only CPU can be used.

Other current open sources detection systems repurpose classifiers or localizers to perform detection.They apply the model to an image at multiple locations and scales. High scoring regions of the image are considered detections.

YOLO use a totally different approach. it apply a single neural network to the full image. This network divides the image into regions and predicts bounding boxes and probabilities for each region. These bounding boxes are weighted by the predicted probabilities.

YOLO model has several advantages over classifier-based systems. It looks at the whole image at test time so its predictions are informed by global context in the image. It also makes predictions with a single network evaluation unlike systems like R-CNN which require thousands for a single image. This makes it extremely fast, more than 1000x faster than R-CNN and 100x faster than Fast R-CNN. See our paper for more details on the full system. Compiled with OpenCV and CUDA , it can make use of Nvdia GPU to achieve high speed image recognition and detection

Haar-Cascade (Image recognition method)

Haar-like features are digital image features used in object recognition. They owe their name to their intuitive similarity with Haar wavelets and were used in the first real-time face detector.

Viola and Jones adapted the idea of using Haar wavelets and developed the so-called Haar-like features.

A Haar-like feature considers adjacent rectangular regions at a specific location in a detection window, sums up the pixel intensities in each region and calculates the difference between these sums. This difference is then used to categorize subsections of an image. For example, let us say we have an image database with human faces. It is a common observation that among all faces the region of the eyes is darker than the region of the cheeks. Therefore a common Haar feature for face detection is a set of two adjacent rectangles that lie above the eye and the cheek region. The position of these rectangles is defined relative to a detection window that acts like a bounding box to the target object (the face in this case).

In the detection phase of the Viola–Jones object detection framework, a window of the target size is moved over the input image, and for each subsection of the image the Haar-like feature is calculated.

The key advantage of a Haar-like feature over most other features is its calculation speed. Due to the use of integral images, a Haar-like feature of any size can be calculated in constant time (approximately 60 microprocessor instructions for a 2-rectangle feature

Results

We decided to only use OpenCV and its built-in Haar Cascade method.

Compiling Opencv on c++ and using the video from the video receiver we are able do real time object detection with relatively low latency.

Thou it is the most early methods and least accurate ,it takes up the least amount of processing time and it is relatively easy to setup.Reason we didn’t use tensorflow to train a specific model is because that requires 1000+ picture of the dummy and we don’t have the picture of the dummy.The next step from here will probably be making use of the YOLO method with GPU to offer a off-board solution.

The Receiver used to connect the computer to fpv cam is a High Quality Eachine ROTG01 UVC OTG 5.8G 150CH.

[mdp-blog2019]

Mission Planner

Mission planner Autopilot

Mission Planner is a full-featured ground station application for the ArduPilot open source autopilot project.

In picture below we can see the location of the flight controller  on the google map with its state of tilt shown in the user interface .With this we know where the drone is and its current state.

Mission Planner Features are detailed in each of the following sections.

The sections are organized to match the major section of the Mission Planner as selected in the menu along the top of the Mission Planner window.

Connect (Upper right corner) - How to connect the Mission Planner to your ArduPilot. Selecting communication devices and rates.

Flight Data - Information about what you see, and things you can do in the Flight Data screens.

Flight Plan - Information about the various aspects of preparing flight plans (Missions).

Initial Setup - Information about what you see and things you can do in the Initial Setup screens.

Configuration Tuning - Information about what you see and things you can do in the Configuration/Tuning screens.

Simulation - How you can use the Mission Planner and a flight simulator to ‘simulate’ flying.

Terminal - Information about what you see and things you can do in the Terminal screens.

Help - About the help screen, and how to get help with your questions about Mission Planner. 

[mdp-blog2019]

PID and Auto-Tuning of drones

What is PID?

A proportional–integral–derivative controller (PID controller or three-term controller) is a control loop feedback mechanism widely used in industrial control systems requiring continuously modulated control. A PID controller continuously calculates an error value as the difference between a desired setpoint (SP) and a measured process variable (PV) and applies a correction based on proportional, integral, and derivative terms (denoted P, I, and D respectively), hence the name.

Correction Capabilities

Tuning– The balance of these effects is achieved by "loop tuning" to produce the optimal control function. The tuning constants are shown below as "K" and must be derived for each control application, as they depend on the response characteristics of the complete loop external to the controller. These are dependent on the behaviour of the measuring sensor, the final control element (such as a ESC), any control signal delays and the process itself. They are normally refined, or tuned, by "bumping" the process in practice by introducing a setpoint change and observing the system response.

PID Tuning in Drones

To simplify how PID rates works in Quadcopter, imagine the drone’s desired tilt to be its goal/set-point. The input power varies based on the drone current tilt relative to the setpoint using the PID function.

There are 3 values in a PID equation, they are the P term, I term, and D term:

“P” looks at present error –  the further it is from the set-point, the harder it pushes

“D” is a prediction of future errors – it looks at how fast you are approaching a set-point and counteracts P when it is getting close to minimize overshoot

“I” is the accumulation of past errors, it looks at forces that happen over time; for example if a quad constantly drifts away from a set-point due to wind, it will spool up motors to counteract it. 

AutoTuning

The flight controller tune itself by twitching around it’s 3 axis repeatedly each time with different PID values until it gets the optimal one. How this achieve the optimal PID rates is by introducing its own disturbance and slowly increasing the P value till its settling time and oscillations are at optimal level .After that it will adjust the D value until the oscillation time is at optimal level. The I value is used in the case of external factors such as wind ,where the user have to set themselves.This process is used to tuned the drone response rate to yaw,pitch and roll control input.

[mdp-blog2019]

Sp Aero 2 (Jo, Yu Fan, Royden) Intelligent payload system

Intelligent Payload

The intelligent payload system, attached with a dual camera will be launched from the mothership the moment the mothership detects and locks onto the target. The long radio system allows the Intelligent Payload to receive guidance signal from an extremely long distance and land on the target.

Flight Controller

Hobbywing Micro

Battery

4-6S (14.8-22.2V)

Radio System

Low Frequency Long range radio (433 MHz) transmission

FPV System

Dual camera, analog and digital

Frame

EMax Carbon fibre frame

Motors

T-Motors F60 Pro V2 1900kV

5-inch Poly Carbonate Propellers

ESCs

Hobbywing 40A 4-in-1, digital signal

[mdp-blog2019]

SP Aero 2 (Jo, Yu fan, Royden) Mothership

SP aero 2 won Second runner up prize after a fierce competition against other international teams.

During this competition, SP aero 2 manage to reach the target using a dual-device system. The team used a concept of a mothership and an intelligent payload system to do Search and rescue for the competition. The drone is fully made from scratch with the help of our team leader Hong Phu, whom he manage to get the custom built aluminium and carbon fiber frame.

The design of the drone was inspired by the many problems encountered by Search and Rescue drones. This is because in search and rescue situations, the drone needs to get up close to examine the condition of the injured and when dropping the first aid kit (payload) the drone cannot be too high as the payload may hit the injured as precision is compromised when the payload is dropped from a great height.

SP aero 2 planned to use a smart intelligent payload instead where it will take off from a mothership once the camera in the mothership is able to detect any people on the ground, the smart payload delivery will then examine the injured and the payload drop decision will then be made by the pilot if it is safe. For the smart payload delivery, the payload will be delivered safely to the injured where it will land near the injured where it can be reached conveniently without any struggles by the injured.

Mothership

The mothership is provided with a 400 cm^2 area of launching space, the mothership is able to carry and launch the intelligent payload system autonomously. With a 16mm telescopic camera, the mothership can also scan the area for casualties and detect the target accurately with the assistance of OpenCV

Flight controller

Pixhawk 4

Battery

6 cell 22.2V

Radio system

2.4 GHz

FPV system

5.8 GHz, dual cameras, switching circuit

Frame

Custom-made frame using carbon fibre and 7075-T6 Aluminium

Motors

T-motors 5212 420kV

18-inch carbon fibre propellers

ESCs

T-Motors Flame 70A

Flight Controller ( Pixhawk 4 )

The Pixhawk 4 is the flight controller used for our mothership.

Pixhawk 4® is an advanced autopilot designed and made in collaboration with Holybro® and the PX4 team. It is optimized to run PX4 version 1.7, suitable for academic and commercial.

We decide to choose Pixhawk 4 as it is the latest version of its series .The Flight controller comes with its own Power Distribution Board and GPS module that are essential for the drone to operate in autopilot more

Using it with Px4/ArduPilot Mission Planner ,we can set a specific path on a geoimage from google map that the drone will follow autonomously .Since we are only limited to one active pilot ,the mothership has to be autonomous when the small human piloted drone/payload is deployed to perform payload delivery.Thus the Pixhawk autopilot capability as a flight controller is very useful for this drone in its search and rescue mission

How the ArduPilot Mission Planner work will be discussed more in the later page of the report.

Aside from its autopilot feature , it also has the ability to autotune its PID rates (Proportional-Integral-Derivative) used to adjust the drones roll,pitch and yaw relative to its 3 axis

Pixhawk Specs

Main FMU Processor: STM32F765

32 Bit Arm® Cortex®-M7, 216MHz, 2MB memory, 512KB RAM

IO Processor: STM32F100

32 Bit Arm® Cortex®-M3, 24MHz, 8KB SRAM

On-board sensors:

Accel/Gyro: ICM-20689

Accel/Gyro: BMI055

Magnetometer: IST8310

Barometer: MS5611

GPS: ublox Neo-M8N GPS/GLONASS receiver; integrated magnetometer IST8310

Interfaces:

8-16 PWM outputs (8 from IO, 8 from FMU)

3 dedicated PWM/Capture inputs on FMU

Dedicated R/C input for CPPM

Dedicated R/C input for Spektrum / DSM and S.Bus with analog / PWM RSSI input

Dedicated S.Bus servo output

5 general purpose serial ports

3 I2C ports

4 SPI buses

Up to 2 CANBuses for dual CAN with serial ESC

Analog inputs for voltage / current of 2 batteries

Power System:

Power module output: 4.9~5.5V

USB Power Input: 4.75~5.25V

Servo Rail Input: 0~36V

Weight and Dimensions:

Weight: 15.8g

Dimensions: 44x84x12mm

Other Characteristics:

Operating temperature: -40 ~ 85°c

Motor Specification

Motors and propellers are chosen to lift approximately 5.5 kg of weight.Since we are using 4 motors ,with the motors and propellers we can have the drone hover at 35% throttle .This mean the motors and propeller are more than powerful enough to lift the drone and its payload.

[mdp-blog2019]

SP Aero 1 (manish)

SP Aero 1, although with good effort by the team, failed to qualify in the first round of the competition due to technical problem with the drone. The nature of which would be described as a material failure of one of the motor mounts that connect to the frame. After analyse of video footage and data of flight recordings this was determined to be the case. The resulting crash cause the drone to suffer large scale damage to the props, GPS mount, remaining arms. Small amount of damage incurred by electronics.   

The evening of the crash saw the team of SP Aero 1 fully rebuild the drone back to operational state. With innovative fixes to the damage by backup spare parts such has disassembling a damaged arm to use its motor mount and replacing the arm with a spare fiber arm tube. The rebuilding process carried on till late in the evening and was concluded in a short test flight.

On the following day, the drone looked to perform the previously assigned task as a proof on concept and operational readiness. Unfortunately, due to what has been theorized to be a total power failure the drone crashed somewhere in the challenge area. The crash, outside the take-off area, resulted in a unsuccessful search to locate the drone. Additionally, it was the drone was described to have “unstable flight” shorty after being directed to the 1st waypoint.

Although the crash was an unfortunate event, the Team SP Aero 1 demonstrated a high level of determination and commitment during the 3-day competition. Additionally they demonstrated good effort and teamwork by additional effort to repair the drone. Overall we believe that both teams demonstrated good cooperation and camaraderie during the length of AAVC.  

The drone was comprised of a standard tarot frame with retractable landing gear. The drone was built for high conventional effectiveness in the Search and Rescue challenge scenario.

Flight controller

Pixhawk (Cube) 2.1

Battery

6 cell 22.2V x 2

Radio system

900 MHz telemetry

FPV system

2.4 GHz, Immersion RC, Ground Station TBS

Frame

Standard Frame Q4 made by Tarot

Motors

T-motors 5212 420kV x 4

18-inch carbon fibre propellers x4

ESCs

T-Motors Flame 70A

Camera

GoPro Camera Hero 5

Flight controller 

We chose to use the industrial pixhawk 2.1 because of its effectiveness as an industrial controller. Modular design for flexibility multiple GPS and Triple redundancy IMU system. Additionally a Isolated, dampened and temperature controlled IMU.

Radio system 

900 MHz for deeper radio penetration over standard 2.1MHz, 433MHz. 

[mdp-blog2019]

Autonomous Ariel Vehicle Challenge 2018

In the period of 21st november to 25th november, team SP aero 1 and aero 2 flew to chiang mai, thailand to participate in this competition.

SP aero 2:

Jo Ravin 

Le Hong Phuc

Royden teh 

YaoJi 

Yu fan

SP aero 1:

Joseph

You Heng

Manish

Jerriel

David.

The theme for the AAVC 2018 in Chiang Mai, Thailand is SAR (Search and rescue).

There will be a take off point where the drone will fly to the disaster zone roughly 300m away to deliver a payload (Med-Kit) to the injured.

[mdp-blog2019]

ROBOTIC LION DANCE

In early september, we were tasked to mimic a lion dance performance using robots and we  purchased Jimu robot from amazon build it. We did some minor customisation to make the robot have better balance and looks more alike to a lion. Jo made some customised movement using Jimu app to program for the performance to mimic an actual lion dance performance. The robot was then showcased at community centres to ministers visiting.

Video Demo             

  Hardwares

The JIMU Robot Inventor Kit has over 600 interlocking, interchangeable parts to build any one of 6 pre-designed animals. The 16 individually programmable servo motors allow you to create stunning movements using Blockly code or with the PRP (Pose, Record, Play) function. However you play, this JIMU Robot kit will provide invaluable hands-on STEM learning and endless programming fun.

Includes 675 snap-together parts, 16 smooth motion robotic servo motors, 1 main control box, 1 lithium-ion battery, power adapter, and quick start guide.

Servomotor:

Each servo has an Identification number to distinguish it from other servos, it also has rotatable rudders and each servo has 3-pin ports. Energy and information can be transmitted between the Main control box and servos.

Connections (Wireless):

Wireless is a term used to describe telecommunications in which electromagnetic waves (rather than some form of wire) carry the signal over part or all of the communication path. Some monitoring devices, such as intrusion alarms, employ acoustic waves at frequencies above the range of human hearing; these are also sometimes classified as wireless.

Bluetooth Compatibility: Bluetooth 4.0

The key new feature of Bluetooth 4.0 is its low-energy technology. This lets device manufacturers replace proprietary sensor technology with Bluetooth, which is a more widely adopted standard. If these same devices had Bluetooth 4.0, they could speak to any Bluetooth 4.0 device, be it phone or computer.

Batteries (Rechargeable Lithium Ion Battery):

A lithium-ion battery or Li-ion battery (abbreviated as LIB) is a type of rechargeable battery in which lithium ions move from the negative electrode to the positive electrode during discharge and back when charging. Li-ion batteries use an intercalated lithium compound as one electrode material, compared to the metallic lithium used in a non-rechargeable lithium battery.

Lithium-ion batteries are common rechargeable batteries for portable electronics, with a high energy density, tiny memory effect and low self-discharge. LIBs are also growing in popularity for military, battery electric vehicle and aerospace applications. (Wikipedia, 2018)

 Software

Jimu Robots uses drag and drop programming app to control the individual servo motors simultaneously or individually.

Drag-and-drop interlocking blocks of customizable sequences as shown below are used to create the Lion Dance.

 In addition to that are also pre-programmed sequence in the app itself made easy for users to interact with the robot without programming themselves.

Each button represent a sequence of movementButton represent sequence of actions also can be created user themselves

[mdp-blog2019]

Overview

MDP ,short for Multi Disciplinary Project, is one of the modules Engineering Academy/EA Programme. In this module ,students are to work with students from different faculties of engineering and engage in engineering projects.

[mdp-blog2019]

Our  team consists of two DASE (Diploma of Aerospace Electronic) students and two DME (Diploma of Mechanical Engineering) students.

Liu Yu Fan

Jo Ravin

Royden Teh

Manish