Received:

• RPi
• RFID system

Ordered:
-Drone electronics, FC, RTK, etc.

• Electronics for charging circuit

Order:

• SSD and SSD Hub

Motion planner's position planning works.

Problem

• Currently the angle is calculated from the start angle to the end angle, but the angle should actual be from start angle to the direction of path angle so the drone flies forward instead of sideways or something. So the drone should turn and fly forward.

SETUP PX4 (Pixhawk 6X) → RPI 4B Ubuntu ROS connection:
https://docs.px4.io/main/en/companion_computer/pixhawk_rpi.html (MAIN ONE)
https://docs.px4.io/main/en/companion_computer/pixhawk_companion.html#serial-port-setup

UART MAVLINK

https://discuss.px4.io/t/controlling-pixhawk-6c-with-raspberry-pi-4b/30712/8?u=marnonel6
I had connection problems with the Pixhawk 6c too. In my case, I was able to solve them by setting
• MAV_1_RADIO_CTL to disabled and
• MAV_1_FLOW_CTRL to force off
See here 13 for the complete description.
THIS PARAMETER DOES NOT SHOW UP!!!sudo
UXRCE_DDS_CFG = 0 (Disabled) (https://github.com/PX4/PX4-Autopilot/releases/tag/v1.14.0-rc1) In a pre-release

sudo mavproxy.py --master=/dev/serial0 --baudrate 57600

Outcomes

I can see the connection over UART is working

Problem

Currently the function path_dubins_cc.serializePath(3); does not change the actual path, but creates a string that can be saved to .csv.

Soluition

Choices:

• Make a PR that changes the original Path
• Write some logic to skip steps if it is to close. Reason for this is there is A LOT of points on a turn making a turn VERY slow.

Connection Guide

Setup Guide:

• Holybro PX4 guide
• PX4 info

Position and orientation works with a /pose visualization in RVIZ, but the coordinates are not 100% correct yet.

• Pegasus docs: https://pegasus.mihr.io/
• Github: https://github.com/PegasusDrone

Davin is designing this in OnShape.

https://www.linkedin.com/posts/fixar-global_fixar-dronetesting-uavs-activity-7097244036175937536-ZooO?utm_source=share&utm_medium=member_ios

Currently only checking if the 3D position is in an Euclidean tolerance.

Add

Check if yaw is in tolerance [Deg] before sending next waypoint.

Problem

Currently when you specify the drone to fly to a position the drone will try and fly there as fast as possible.

Solution

Use a flight path generator like an agriculture field cover/ lawn mover generator to generate a flight path to cover a field.

Resources

https://github.com/Fields2Cover/Fields2Cover
https://github.com/Fields2Cover/fields2cover_ros

Load sharing circuit
Charge LiPo battery while by passing power from power source to RPi to keep it awake

Problem

When setting off-board control and arming the drone the drone does not always get armed and then the code just hangs.

Solution

Add code that reads the drone stage and if the drone was not armed and set to off-board then do it again. If this does not work for X amount of tries then Error and stop everything.

Descretization step increased to 0.3m resulting in a 1m/s ground speed in SITL Gazebo simulation.

Requirements:

• Drone without payload 5kg @50% throttle/thrust
• Flight time 35-45min
• Payload 2kg with a minimum flight time of 20min with throttle <=80%
• Space underneath drone for package delivery system

Possible improvement parts:

• Motors -> ESC -> Battery
• Frame change to a HexaCopter (6 Motors - Can maybe use the current motors and buy two more)
• New flight computer VOXL2, Pixhawk 6C, 6X or 6X with RPi compute module.
• NVIDIA Jetson Orin™ NX 16GB -> Will have support for 22.04 in Jetpack 6 release in Q3 2023.

TODO

• Create take-off position -> Current possition and 5m altitude

• Set take-off angle to be current drone angle

Problem

Arming and off-board control does not always work.

Current implementation

Wait some time (1second I think) then arm and switch to off-board control.

Proposed solution

Wait 1 second then arm the drone and subscribe to the drone mode status and see if the drone is armed and that it is off-board mode if it is not then wait 1 second again and retry. Have a max number of retries say 5 then if it still is not armed then stop everything and indicate that drone does not want to arm.

RPi as High level controller with MavLink

Username: Pegasus
Password: ?

Xavier Running:

• Ubuntu 20.04
• ROS2 Galactic - With a ROS bridge

Upgrade to:

• NVIDIA Jetson Orin™ NX 6-16GB -> Will have support for 22.04 in Jetpack 6 release in Q3 2023.

• Old software.
• Do not see his setup or parameters anywhere.

Full electronic schematic of drone and charging hub

• High level connections: Inkscape with photos
• Electronic design of charging circuit and PCB: KiCad

Problem

I am trying to tune my drone's PID controllers indoors in a custom rig that I built. I have PX4 v1.14-rc, but I keep getting the error in the images below. I have all the sensors calibrated.

I can't arm the drone and the M8N GPS is flashing red on the drone's fast and continuously. I think that is just because it does not have a GPS lock, because I'm indoors.

When I try to arm the drone with the remote control it only makes a sound, but does not arm.

I am building and uploading the banch release/v1.14 and I tried release/beta -> v1.14.0-rc2

Problems

• PX4 says negative Z is up where ROS is positive Z is up
• Add local coordinates [NED] to GPS coordinates function
• Here is ROS PX4 frame conversions

Currently the drone has unstable flight. It Yaw's continuously and drifts in pitch and roll.

Problem with community help

PX4 forum post with logs

Dronecode foundation Discord chat:

Key points from suggestions:

• Mechanical vibrations/resonance introduced by loose landing gear or soft flight controller dampers
• PID not tuned at all

Suggested solutions

• Cable tie around FC dampers to make it more rigid
• Make sure no cable is to stiff on the FC and pulling it
• Tune PID gains

• Increase proportional gain for yaw and roll rate controller

USE RTL it works now

Add a function for path discretization between any two points in a straight line. This will help with flying the drone to and from the start and end of the Fields2Cover path and also back to home or takeoff smoothly.

Use pre-release PX4 v1.14 and the new DDS with (Ignition) Gazebo.

Batteries:

1. 6S - Charged - Health looks good
2. Radiomaster Li-Ion 7.5V 5Ah 37Wh 5000mAh 2s (Transmitter) - Ordered new battery

Install and setup YOLOv8 training environment for Ubuntu 22.04 and CUDA >=12.1

• Official Quickstart

Steps:

1. pip install ultralytics
2. pip3 install torchvision==0.16
3. pip3 install torch==2.1
4. pip3 install torchaudio

CLI YOLO interface

Python interface

# Train with Python
from ultralytics import YOLO

# Load a model
model = YOLO('yolov8n.pt')  # load a pretrained model (recommended for training)

# Train the model
results = model.train(data='<path_to_dataset>/data.yaml', epochs=300, imgsz=640)

Get path from fields to cover then command drone to fly that path in Gazebo with SITL

Error

/home/marno/Classes/Final_project/Drone_feedlot_precision_farming/osprey/src/precision_farming_drone/drone_control/src/path_planning.cpp: In member function ‘void PathPlanning::path_planning()’:
/home/marno/Classes/Final_project/Drone_feedlot_precision_farming/osprey/src/precision_farming_drone/drone_control/src/path_planning.cpp:153:43: error: ‘using F2CPath = struct f2c::types::Path’ {aka ‘struct f2c::types::Path’} has no member named ‘discretize_swath’
153 | F2CPath new_path = path_dubins_cc.discretize_swath(discretize_step_size);
| ^~~~~~~~~~~~~~~~
gmake[2]: *** [CMakeFiles/path_planning.dir/build.make:76: CMakeFiles/path_planning.dir/src/path_planning.cpp.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:165: CMakeFiles/path_planning.dir/all] Error 2
gmake: *** [Makefile:146: all] Error 2

1. Balance drone weight

• Thrust is not enough. @50% throttle only 4kg of thrust - drone is currently 2.3kg with a flight time of 28min
• Increasing the weight will drastically decrease the flight time. @5kg the drone will be at 85% throttle just for hovering with a flight time of <10min probably.