Industrial Departement

Graduation Internship
To Obtain the State Engineer Degree in :

Mechatronics Engineering

Development AND Simulation of an

AUTONOMOUS DRONE
Defended on 24/06/2024

Realized by : Supervised by : The honorable Jury :

• Pr. KAMMOURI ALAMI SALAHEDDINE • Pr. KAMMOURI ALAMI SALAHEDDINE
EL HARRACHY MERYEM
• Mr. AIT MOUMMAD ABDELKRIME • Pr. EL OUAZZANI NABIH
• Mrs. TANTOUI ELARAQI SAFAA • Pr. DRISS TAHRI
PLAN

Problematic & Solution

Objectives

Drone Control

Drone Path Planning

FCND Simulation

Conclusion & Perspectives

The company overview
2
Capgemini engineering is a global information technology service company that offers
engineering and technology consulting solutions.

Host Department MBSE & System Engineering

 Problematic
3
Manual remote control
 Traffic congestion
 High logistical costs
 Limited access to remote
areas
Objectives
4

01 02 03

Drone Path FCND

Control Planning Simulation

Implement control Develop planning Test and visualize the

algorithms to ensure algorithms to optimize route results in a virtual
precise and stable conditions, considering time, environment.
trajectory maintenance for distance, and accessibility
the autonomous drone. constraints
Objectives
5
Drone Control 6

Drone
01
Control
Drone Control
7

Up and Down movement of the drone Thrust [N]: U1

Forward-Backward movement of the drone Roll Moment [N.m] : U2

Left- Right movement of the drone Pitch Moment [N.m] : U3

Allows the drone to Pivot in place Yaw Moment [N.m] : U4

 The Control System Architecture
Drone Control
8
Control the drone Position & Attitude

The desired Trajectory that the drone should achieve

Simulate the drone
 The Control System Architecture
Drone Control
9

Two subsystems:
 Force subsystem which comprises the translational equations
 Moment subsystem which comprises the angular equations

UAV system structure

 The Control System Architecture
Drone Control
10
 Attitude Controller
Drone Control
11
MPC (Model Predictive Control)
An advanced control algorithm that uses a dynamic model of the system to predict its behavior according to flight
conditions and adjust propeller movements in real time to maintain in-flight stability.
 Attitude Controller
Drone Control
12
MPC (Model Predictive Control)

An advanced control algorithm that uses a dynamic model of the system to predict its behavior according to flight
conditions and adjust propeller movements in real time to maintain in-flight stability.

MIMO Constraints tolerance Preview Capability

Multi-input Multi-output systems
 Attitude Controller
Drone Control
12
MPC (Model Predictive Control)

An advanced control algorithm that uses a dynamic model of the system to predict its behavior according to flight
conditions and adjust propeller movements in real time to maintain in-flight stability.

MIMO
Multi-input Multi-output systems Constraints tolerance Preview Capability
 Attitude Controller
Drone Control
12
MPC (Model Predictive Control)

An advanced control algorithm that uses a dynamic model of the system to predict its behavior according to flight
conditions and adjust propeller movements in real time to maintain in-flight stability.

Constraints tolerance Preview Capability

MIMO
Multi-input Multi-output systems
 Attitude Controller
Drone Control
12
MPC (Model Predictive Control)

An advanced control algorithm that uses a dynamic model of the system to predict its behavior according to flight
conditions and adjust propeller movements in real time to maintain in-flight stability.

Preview Capability
MIMO Constraints tolerance
Multi-input Multi-output systems
 Attitude Controller
Drone Control
13

Choose control inputs that minimize this cost function while satisfying system constraints
 The Control System Architecture
Drone Control
14
 Position Controller
Drone Control
15
 Position Controller
Drone Control
15

 Measure errors:
• Position / Velocity Error: difference between desired and actual drone
position / velocity.
 Calculate corrective acceleration:
• Differentiate velocity error to find acceleration needed to correct errors.
 Control adjustments:
• Continuously adjust drone controls based on calculated acceleration.
 Maintain stability
Drone Control
16

Simulation and Results

 Simulation and Results
Drone Control
17
Spiral trajectory

• Spiral trajectory radius : 2 meters

• Initial height : 2 meters
• Final height : 30 meters
• Initial rotational velocity of all rotors: 150 rad/s
• Initial coordinates of the UAV:
x=0 meters
y=-1 meters
z=0 meters
 Simulation and Results
Drone Control
18

Spiral trajectory

Strong overshoot at the beginning of the test period.

That can be explained by the fact that the drone
starts its journey from quite a long distance away
from the trajectory. However, once it reaches the
path that it needs to follow, the velocities of the UAV
stabilize and track the reference values very
smoothly.
 Simulation and Results
Drone Control
19
Positions & Velocities values as a function of time
 Simulation and Results
Drone Control
20

Propeller’s velocities as a function of time. Force & Moments as a function of time.

Path Planning
21

Path
02
Planning
Path Planning
22
Path Planning
Trajectory planning or route planning, is the process of determining an optimal path or route for an entity to travel
from a starting point to a destination .
Path Planning
22
Path Planning
Trajectory planning or route planning, is the process of determining an optimal path or route for an entity to travel
from a starting point to a destination .
Path Planning Different methods of Path Planning
23

Classical Approach

The AI Approach
Path Planning Different methods of Path Planning
25
Dijkstra A* D* D* Lite RRT

Static Dynamic
Environnent Static Dynamic Dynamic

Optimality (1-10 scale) 10 9 8 8 6

Speed (Time to Compute, ms) 500 200 300 250 150

Consideration of Limited High Very high

Limited High
movement dataset

Robustness to
Low (2D) Modium (2D/3D) High (3D) Very high (3D) Very High (3D)
dimensionality
Path Planning
26

D* Lite Process & Results

Path Planning D* Lite Process
27
Path Planning D* Lite’s Results
28
Path Planning
29

Path Planning using Deep Learning

 Path Planning Path Planning using Deep Learning
30

Convolutional Neural Network (CNN) inspired by the U-NET architecture, is a specialized type of deep learning
algorithm mainly designed for tasks that necessitate object recognition, including image classification, detection, and
segmentation
 Path Planning The Dataset of the Model
31
START TRAJECTORY

2D Path Planning Dataset: Created from randomly generated

100x100 pixel maps with varying levels of obstacle density, this
dataset is designed for path planning tasks in a controlled
environment.
• 170k for training
• 50k for testing
• 15k for validation
OBSTACLE GOAL

NB: The ground truth paths are generated using D* lite algorithm
Path Planning
32
TRAINING

Optimizer Schedular Loss Function

Cosine Annealing
Binary Cross-
• Your text here
ADAM w i t h Wa r m
• Your text here Entropy
• Your
R etext
s there
arts
• Your text here
Path Planning
33

45 EPOCHS

An Epoch represents one iteration

over the entire dataset
Path Planning The Training Results
34
Path Planning The Training Results
34

95% 40%

Accuracy Precision
FCND Simulation
35

03 FCND SIMULATION
FCND Simulation The Necessary Tools
36
FCND Simulation State machine of the Model
37
FCND Simulation The Simulation Results
38
Conclusion Future Perspectives
39

o Improve obstacle detection using advanced sensors (LiDAR, cameras, ultrasonic sensors, radar)

o Enhance adaptability to dynamic environments.

o Application of Reinforcement Learning:

 Enable the drone to learn and adapt its behavior based on environmental feedback.
 Optimize performance over time.
Thank You !!