This is the official and actively maintained RosettaDrone project.
RosettaDrone is a framework for developing and testing software for DJI drones. It provides a MAVLink wrapper which allows users to control DJI drones using MAVLink-speaking ground control stations and offers tools and features for implementing and testing AI scripts.
*** IMPORTANT SAFETY NOTE ***
Always stay away from the propellers. Use RosettaDrone's "Safe Mode" to prevent the drone from unexpectly arming or taking off. When testing a new Rosetta version, test it first using the simulator.
This Software is experimental and under constant development. By using this Software, you agree to assume all responsibility for eventual harm or damage. See "Known issues" below before use.
OpenSource developers are welcome. If you use this Software, we expect you to get directly involved with the source code, read the issues section and, most important, contribute with pull requests.
This new version implements a MissionManager that speaks native MAVLink and uses VirtualSticks. This enables waypoint missions on DJI Mini series and other drones without "on-board" waypoint missions support. The MissionManager can also be used with DJI drones that support WayPoints and MissionControl onboard. This can be an interesting alternative for improving the features offered by their flight controllers.
Rosetta has been tested on:
- DJI Mini
- DJI Mini SE
- DJI Mini 2
- DJI Mini 2 SE
- DJI Air
- Mavic 2
- Mavic Pro
- Mavic AIR series
- Matrice 210 V2
Rosetta is compatible with:
- QGroundControl
- Mission Planer
- ArduCopter SITL
- Dronekit-Python (AI scripting)
- MavProxy
- MAVSDK
- DJI SDK 4.16.1
- If you want to contribute please follow these important recomendations.
- Here is some documentation from a forked project.
- Rosetta can be tested without a Drone (press 5 times on the logo). For MAVLink and GUI testing.
The first time you start the app it will attempt to connect to DJI with the provided API key.
This version implements the new video stream method in DJI SDK that require quite a few modifications. It also includes all the latest libraries, and Androidx. The DJI Simulator is enabled in the unstable code for testing in the lab.
DJI-Rosettadrone-QGC live over 4G-LTE
Main screen, with live video on Mavic AIR (from my office)...
Main screen on QGroundControl with live video and telemetry... (from inside my office)
RosettaDrone on the big screen from DJI Smart Controller
RosettaDrone on the DJI Smart Controller
Latency video: (https://youtu.be/HAC5OpQQpDI "Click to play in Youtube.com")
- Report telemetry in QGC like position, attitude, relative altitude, heading, and battery remaining
- Command Return-to-Launch from QGC
- View drone video feeds in QGC or forward RTP to an IP address of your choice (currently Mavic Pro 2 only)
- Create and fly waypoint missions (with and without VirtualSticks)
- Fly by joystick and QGS
- Fly from Python in DroneKit
- Use Mavproxy to connect both QGC and DroneKit at the same time
- Use Gstreamer/OpenCV/FFMPEG and DroneKit to create complex AI functions
-
Connect your Android phone to a DJI transmitter and power on your DJI drone.
-
Start RosettaDrone. The DJI light in the top-right will turn green if the app is successfully communicating with your drone.
-
If you wish to use QGroundControl on an external device, click the Gear icon to access Settings, check Use GCS on an external device, then specify an IP address. Additionally a second telemetry connection con be configured to run AI scripts using DroneKit Python.
-
Start QGroundControl. A telemetry connection should be immediately established, and the GCS light in RosettaDrone will turn green. Note that if you are using QGroundControl on the same device as Rosetta drone, the GCS light may not turn green if QGC is in the background.
-
To start video:
a. Click the "Q" icon in the top-left corner of QGC
b. Under Video change Video Source to UDP Video Stream.
c. Change UDP Port to 5600.
-
Takeoff is recommended using the RC transmitter. To arm or takeoff from the GCS, click the SAFETY ENABLED button. It will turn green and say READY TO FLY. Then use the QGroundControl Takeoff or Start Mission function. Takeoff and missions should also work fine using QGC.
-
After flight, ensure Safe Mode is enabled before approaching propellers.
-
Support is added for Pan and Tilt of camera gimbal by using RC channel 8 and 9.
def set_servo(servo,val): msg = vehicle.message_factory.command_long_encode( 0, # time_boot_ms (not used) 0, # target system, target component mavutil.mavlink.MAV_CMD_DO_SET_SERVO, 0, servo, # RC channel... 1500+(val*5.5), # RC value 0, 0, 0, 0, 0) vehicle.send_mavlink(msg) vehicle.flush()
And then use: set_servo(8,-45.0) to set the gimbal....
-
Support for joystick has been added and tested with QGroundcontrol, using the now built in simulator.
-
By tapping on the drone icon 5 times you enable test mode and can open the software with no drone connected. Useful for testing MAVLink communication and GUI.
-
Install git lfs as specified at https://github.com/git-lfs/git-lfs/wiki/Installation
-
Clone or download the repository.
-
In Android Studio, select File->New->Import Project and navigate to the downloaded folder.
-
Sign up for the DJI Developer Program at https://developer.dji.com/mobile-sdk/ and create an Application key. The package name should be sq.rogue.rosettadrone.
-
Generate Google Maps API key using instructions at https://developers.google.com/maps/documentation/javascript/get-api-key#creating-api-keys
-
Create a new file called keys.xml in the /app/src/main/res/values/ folder, and insert the following:
<?xml version="1.0" encoding="utf-8"?> <resources> <string name="dji_key">INSERT KEY HERE</string> <string name="google_key">INSERT KEY HERE</string> </resources>
-
Run Build->Make Project
Anyone who speaks multiple languages knows that translations are rarely perfect. The same is the case here.
-
The ARM button in QGC does not work, by design. Sending a "Takeoff" or "Start Mission" command from QGC will arm the motors and takeoff.
-
DJI reports heading in True, which Rosetta passes along in vfr_hud.hdg. The mavlink protocol does not specify magnetic or true.
-
DJI and Mavlink use different scales to characterise GPS accuracy. DJI also does not report hdop or vdop.
-
Radio signal strength is not yet implemented
-
Rosetta drone reports groundspeed and airspeed as the same, and does no wind correction
-
There is currently no way to turn the video camera on or off in QGC, but DJI controllers will still work fine, Mavlink support for this is now added.
-
The only implemented waypoint actions include delaying at a waypoint, taking a photo, or changing the gimbal pitch
-
Battery % is now show in the GUI on the main screen.
-
Battery Warnings for both Drone and Controller at 20% 10% and 5%. No sound as the moment.
-
Controller stick positions and C1, C2 and C3 switch states are now reported as RC RAW messages over Mavlink
-
A working DJI simulator is added for testing and training. The simulator is Hardware In the Loop (HIL), and this require both a controller and a drone connected.
-
A test mode is added by hitting 5 times on the drone icon. This will enable GUI testing only.
-
The message on the boot/login screen can now be modified in "settings/drone configuration/App Name" , to make tha app more private.
-
The switch C3 is now both reported as switch 3 and also perform an auto takeoff is the system is armed. This to allow flying from RosettaDrone, and might be reverted.
-
If you change the UDP port while QGC is running, you may need to restart QGC.
-
If you change the video UDP port while QGC is running, you may need to restart RosettaDrone.
-
Some times the RosettaDrone app stops and restarts. Work is in progress to eliminate this. Seems to be related to the MAVLink layer.
-
If you takeoff using the QGC, set a point, and then hit goto, RosettaDrone enters an error state. Work on a fix for this is in progress.
RosettaDrone uses MAVLink code generated by the MAVLink code generator, using the ArduPilot dialect. The Java code generator contains errors (see issues #805 and #806), and the code required manual tweaking after generation. This means that simply dropping in updated auto-generated files will likely produce errors. At the moment this seems only to apply for setting and getting double that is missing. It is quite easy to copy the float function and modify it to double. The only usage for this is for ROV wheel distance, so someone should fix that in Mavlink.
The use of an ArduPilot dialect over PX4 is not intended to be a statement of preference. The author believes strongly in the importance of maintaining maximum compatibility across both projects.
RosettaDrone was brought to you by developers from Rogue Squadron, a UAS/C-UAS red team at the Defense Innovation Unit Experimental, 9Tek AS (Norway), IMATRONIX and others.
RosettaDrone is licensed under BSD3. For particularities related to U.S. federal government code release, please read Intent.MD. For more information, visit code.mil or code.gov.
RosettaDrone uses a modified version of DJI's Android Video Stream Decoding Sample, which is released under the MIT License.
Video RTP packing uses code modified from the libstreamer library, licensed under Apache 2.