UAV guided UGV

Presentation: Link

Installation and setup

Create a catkin workspace by running the following commands

mkdir -p catkin_ws/src
cd catkin_ws/src
git clone https://github.com/dalalbhavya/uav-ugv.git
catkin build
source ../devel/setup.bash

To run the world make sure to include the path in $GAZEBO_MODEL_PATH environment variable

src/InterIIT_DRDO/interiit22/models

by running the command in terminal

echo $GAZEBO_MODEL_PATH
export GAZEBO_MODEL_PATH=$GAZEBO_MODEL_PATH:<your path>

To install PX4-Autopilot use the commands

git clone https://github.com/PX4/PX4-Autopilot.git --recursive
bash ./PX4-Autopilot/Tools/setup/ubuntu.sh

ROS Topics not publishing for depth camera Solved by copying the links a new world file drdo_world1_cam.world

P.S. : This isssue is yet to be resolved in a more general way

To perform a general testing of the control code of drone follow

1. Open a new terminal and change the directory to ~/catkin_ws/src/iq_sim/scripts and run

./startsitl.sh

2. Now open a new terminal and run (replace last digit for respective world (e.g. drdo_world2.launch for world2)

roslaunch interiit22 drdo_world1.launch

3. In a new terminal run

rosrun iq_gnc drdo_test

4. Now make the mode GUIDED by running the following command in MAVProxy Terminal

mode GUIDED

---

Approach

Mapping and Geotagging

We wrote code for manoeuvre the UAV over the terrain and capture rgbd images of small patches of terrain and created an orthomosaic map . UAV used Ardupilot MissionPlanner for maneuvering and uses RTL command to return back to start position.

Road Detection

D-LinkNet is used to segment the roads obtained by rgbd camera images We skeletonize the output mask generated by the D-LinkNet model

For case of Plain Overlay images - We use Depth Image based Plane Detection (DIPD) algorithm to distinguish the roads from the terrain .

Image Thresholding and Tracking of Car

We are using image thresholding for determining orientation i.e. yaw of the prius or UGV. We are using PID controller for tracking of the car.

UGV Movement and concerned ROS Topics & Nodes :

We created a ROS node string_generator for converting array of waypoints into a string of commands which is then published to prius_teleop node. We use cross product function for detecting turns.

For 3 consecutive points ( x1 , y1 ) , ( x2 , y2 ) and ( x3 , y3 ) ;

cross_product = V12 x V13 = ( x1 - x2 )( y3 - y2 ) - ( x3 - x2 )( y2 - y1 )

cross_product = 0 —> go straight

cross_product > 0 —> left turn

cross_product < 0 —> right turn

We created a ROS node prius_teleop for utilizing string of commands for generating control msgs for publishing to /prius topic .

/prius -> We use this topic for controlling motion of prius car . We publish values of throttle , steer , brake and shift_gears for maneuvering the car.

Spline Interpolation and Path Planning :

We are using A* algorithm for generating waypoints through which UGV has to transverse to reach target location. We are using Cubic splines for smoothening of turns.

References

1. https://docs.px4.io/master/en/simulation/ros_interface.html
2. https://openaccess.thecvf.com/content_cvpr_2018_workshops/w4/html/Zhou_D-LinkNet_LinkNet_With_CVPR_2018_paper.html
3. https://competitions.codalab.org/competitions/18467
4. https://github.com/Intelligent-Quads
5. https://github.com/OpenDroneMap/ODM
6. https://bdataanalytics.biomedcentral.com/articles/10.1186/s41044-018-0035-y
7. https://ardupilot.org/copter/docs/zigzag-mode.html