-
Notifications
You must be signed in to change notification settings - Fork 344
Launch Sequence and Flying
This page describes how to prepare and fly a drone while running code on Jetson TX1/TX2 companion computer. Make sure you read and understand all of the steps and do a dry run before going out to the field in order to ensure that all of the components are working as expected.
Please make sure that you have a safe environment to fly the drone. In particular:
- Make sure you know and obey all relevant local laws and regulations.
- Make sure you have a way to disable drone motors directly from the transmitter via a "kill switch". Follow the instructions to set up this switch.
- For the very first experiments we recommend flying the drone in an open field where operating RC aircraft and drones is permitted. As an additional safety measure, we suggest tying the drone to the ground using sufficiently long and lightweight rope.
- Make sure the drone flies stably and with no issues before running the code on Jetson and switching to autonomous mode.
There are several steps that need to be done before launch.
- Power on the drone by connecting the battery.
- If you rely on PX4FLOW module for stabilization and visual odometry do the following:
- Wait for Pixhawk's initialization beep and then lift the drone for 20-30cm above the ground.
- Wait for the drone's LED light to turn green (if Iris+ and Pixhawk are used). PX4FLOW is initialized when this happens.
- Put the drone back on the ground
- Close the battery compartment and arm Pixhawk by pressing and holding the arm button. The button's LED light should start producing fast and short blinks. If this happens, the drone is armed. If not, re-check your hardware and Pixhawk's firmware.
Check that the Jetson is powered as well. Wait a minute or two for the Jetson to boot, and connect to its WiFi network. Next, SSH to the board:
ssh nvidia@10.42.0.1
To speed up this login process and avoid entering the password on each login, you can define the following alias:
alias sshtx="sshpass -p 'nvidia' ssh nvidia@10.42.0.1"
This assumes default user/password used on Jetson. Note that in previous versions of JetPack (e.g. 3.0), the default user/password combination was different: ubuntu/ubuntu
. For better security, it is strongly recommended to change the password before going to the field/trail.
Make sure you have the latest source code installed on the Jetson. If you ran the setup script as a part of Jetson setup, then the code should be in ~/redtail
directory. Pull and build the latest version if needed (connect network cable to the Jetson carrier board):
cd ~/redtail
git pull
cd ~/ws
catkin_make
TIP: you can mount Jetson home directory to any directory on the host computer to simplify access to Jetson filesystem by using sshfs
tool:
sudo sshfs -o idmap=user -o allow_other nvidia@10.42.0.1:/home/nvidia /home/[your_directory]/tx2-01/
Once the code is built, there are multiple options to run the ROS components.
There are total of 3 components that run on the drone and ground station. Each component is launched via roslaunch
or rosrun
command.
These include TrailNet DNN, YOLO DNN, MAVROS and camera ROS nodes. To run all of these components, execute the following command from ~/ws
directory on the Jetson:
roslaunch caffe_ros everything.launch
Take a look at ~/redtail/ros/packages/caffe_ros/launch/everything.launch
for more details on what parameters are supported.
To enable drone control with a joystick, connect the joystick (e.g. NVIDIA Shield or Microsoft XBox controller) to your ground station computer and run the joy
ROS node on the ground station:
rosrun joy joy_node _dev:=/dev/input/js0 _autorepeat_rate:=30
You might need to change input device if you have multiple joysticks. Make sure to set the following environment variables (either using export
or via .bashrc
file):
export ROS_MASTER_URI=http://10.42.0.1:11311
export ROS_IP=your_ground_station_computer_ip
ROS_MASTER_URI
should point to machine where ROS master is running (e.g. Jetson) while ROS_IP
is the IP address of the machine on which the joystick node is running (use ifconfig
to determine this address). If you are running the joystick node from a Docker container, make sure the joystick is connected to the computer before starting the container.
NOTE: it is strongly recommended to run the command below when the drone is already in the air and flying in POSITION CONTROL mode. In such case altitude gain should be set to 0 otherwise, the initialization procedure in px4_controller node will try to change the drone's altitude according to the setting which may lead to sudden changes in the altitude.
rosrun px4_controller px4_controller_node _altitude_gain:=0 _linear_speed:=3 _joy_type:="shield" _obj_det_limit:=0.3
Sometimes you may wish to launch components separately. To do that you will need to run MAVROS first via roslaunch
(which will run ROS master as well as the MAVROS node) and then components of your choice via rosrun
commands.
roslaunch mavros px4.launch fcu_url:="/dev/ttyTHS1:921600" gcs_url:="udp://@10.42.0.85"
NOTE: you might need to change IP address in gcs_url
parameter depending on your host WiFi configuration. Run the ifconfig
command to make sure you are using correct IP address otherwise QGroundControl will not be able to connect to the drone.
First, make sure that Jetson board has the camera attached and it is displayed in ls /dev/video*
output.
To enable camera feed publishing in ROS, run gscam
node:
cd ~/ws/
rosrun gscam gscam _gscam_config:="v4l2src device=/dev/video0 ! video/x-raw, width=640, height=360 ! videoconvert"
If you want to enable streaming video to ground station using H.265 encoder, use a different value of the _gscam_config
parameter:
rosrun gscam gscam _gscam_config:="v4l2src device=/dev/video0 ! tee name=t ! queue ! videoconvert ! omxh265enc ! video/x-h265, stream-format=byte-stream ! h265parse ! rtph265pay config-interl=1 ! udpsink host=10.42.0.211 port=6000 t. ! queue ! video/x-raw, width=640, height=360 ! videoconvert"
Change host IP address as needed.
You can also use environment variable instead of parameter to set gscam
GStreamer pipeline:
export GSCAM_CONFIG="v4l2src device=/dev/video0 ! tee name=t ! queue ! videoconvert ! omxh265enc ! video/x-h265, stream-format=byte-stream ! h265parse ! rtph265pay config-interl=1 ! udpsink host=10.42.0.211 port=6000 t. ! queue ! video/x-raw, width=640, height=360 ! videoconvert"
rosrun gscam gscam
To run the TrailNet DNN node, run caffe_ros
node with the TrailNet model:
cd ~/ws/
rosrun caffe_ros caffe_ros_node __name:=trails_dnn _prototxt_path:=/home/nvidia/redtail/models/pretrained/TrailNet_SResNet-18.prototxt _model_path:=/home/nvidia/redtail/models/pretrained/TrailNet_SResNet-18.caffemodel _output_layer:=out _use_fp16:=true
To run the YOLO DNN node, run caffe_ros
node with the YOLO model:
cd ~/ws/
rosrun caffe_ros caffe_ros_node __name:=object_dnn _prototxt_path:=/home/nvidia/redtail/models/pretrained/yolo-relu.prototxt _model_path:=/home/nvidia/redtail/models/pretrained/yolo-relu.caffemodel _output_layer:=fc25 _inp_scale:=0.00390625 _inp_fmt:="RGB" _post_proc:="YOLO" _obj_det_threshold:=0.2 _use_fp16:=true
The steps below describe one possible way of starting autonomous flight, you may have a different sequence of steps if needed.
-
Make sure all components except the controller are running.
-
Take off and switch to POSITION CONTROL mode. Make sure the drone is flying stably (no drift etc).
NOTE: if you don't use GPS, don't have automatic take off and use PX4FLOW module for stabilization and visual odometry you need to follow this procedure for taking off:
- Take off in STABILIZED mode
- Once airborne, switch to ALT HOLD mode and stabilize your drone in horizontal plane manually until the Pixhawk's LED light goes green, which indicates that PX4FLOW module is initialized and working properly.
- Switch to POSITION CONTROL mode.
-
Position the drone in the desired location and orientation.
-
Run the controller node and wait until it switches to NAVIGATE mode (look at the console output).
-
Switch to the OFFBOARD mode from the transmitter. The light on the drone should blink and remain green. If it switches to purple then most likely controller node is not running properly.
-
Check that you can control the drone using joystick:
- left stick:
- up/down: change altitude higher/lower
- left/right: yaw
- right stick:
- up/down: fly forward/backward
- left/right: strafe left/right
- left stick:
-
Position the drone on the trail and press and release A button on the joystick. The drone should start flying along the trail. You can disable autonomous mode and stop the drone using B button. You can also control the drone with the joystick even in autonomous mode - the joystick commands have higher priority than DNN.
In case something is not working, try running each component separately and check if it's working as expected.
Check that the camera is attached and working. Verify that Jetson has recognized the camera by running ls /dev/video*
- you should see at least one video device in the output.
Run gscam
node with H.265 streaming enabled and verify that you can receive the video on the host by running the following on the host:
gst-launch-1.0 udpsrc port=6000 ! application/x-rtp, encoding-name=H265,payload=96 ! rtph265depay ! h265parse ! avdec_h265 ! xvimagesink
You should see the window open with camera feed from the drone.
Next, check that the camera ROS topics are available:
rostopic list | fgrep camera
You should see 2 camera topics: /camera/image_raw
and /camera/camera_info
Finally, check that the ROS topic is being published by running:
rostopic echo -n 1 /camera/image_raw
You should see (a lot!) of output.
To debug DNN nodes, run the gscam
node first and then run the DNN node (TrailNet or YOLO depending on which you are trying to debug) using the appropriate rosrun
command described above. Check whether the DNN node is publishing any data:
rostopic echo /caffe_ros/network/output
You should see the output of the DNN node. If not - check console output as well as ROS logs at ~/.ros/log/latest/
.
NOTE: if you are running DNN nodes using everything.launch
, ROS topic names will be /trails_dnn/network/output
for TrailNet and /object_dnn/network/output
for the object detection net.
YOLO DNN:
- If YOLO DNN node fails with "file not found" error, make sure you have a final model.
- If you are running the YOLO for the very first time, it may take up to 3-5 minutes for YOLO to be compiled and loaded. The compiled model will be cached so it's just first-time delay only.