Obstacle Avoidance in NVIDIA Jetson with PX4 Autopilot in Hardware-in-the-Loop (HITL) Simulation with UAV Dynamics Modeled in Simulink
This example shows how to use the UAV Toolbox Support Package for PX4 Autopilots to verify an Obstacle avoidance algorithm deployed on NVIDIA Jetson as Onboard Computer along with Pixhawk hardware board, in HITL mode and the UAV Dynamics contained in Simulink.
The Obstacle Avoidance block provided as part of UAV Toolbox is an autonomous algorithm that computes an obstacle-free direction using visual sensor data and destination position. For such computationally intensive algorithm, you can use an Onboard Computer on the drone along with the Autopilot. NVIDIA Jetson is a commonly used Onboard Computer for drones. Using Simulink, you can design a complex Autonomous algorithm and deploy the same on NVIDIA Jetson.
This example also showcases 3D scenario simulation during the flight and streaming the image to the onboard computer using the simulated camera sensor. Unreal Engine® simulation environment is used for the 3D scenario Simulation and visualization. This image can be received in NVIDIA Jetson and can be used by the Obstacle avoidance block.
In summary, in this example you,
Deploy an Obstacle avoidance algorithm on NVIDIA Jetson that takes vision data from Unreal and provides path correction to PX4..
Enable flight visualization with PX4 HITL and stream simulated camera image to NVIDIA Jetson.
Run and complete a UAV mission with onboard computer.
Limitation: The Unreal Engine simulation environment is supported only on Microsoft® Windows® system. If you are using a Linux system, skip adding the 3D scenario simulation step and still be able to complete this example.
If you are new to Simulink, watch the Simulink Quick Start video.
Go through the PX4 Hardware-in-the-Loop System Architecture document to understand the physical connections required to setup the Pixhawk and Simulink for HITL Simulation.
Configure and set up Pixhawk in HITL mode. For more information, see Setting Up PX4 Autopilot in Hardware-in-the-Loop (HITL) Mode from QGroundControl.
Setup the PX4 Firmware as mentioned in Set Up PX4 Firmware for Hardware-in-the-loop (HITL) Simulation. In the Select a PX4 Autopilot and Build Target screen, select any Pixhawk Series board as the PX4 Autopilot board from the drop-down list. This example uses Pixhawk 4.
Familiarize with the co-simulation framework for UAVs, see Unreal Engine Simulation for Unmanned Aerial Vehicles
It is recommended to understand the PX4 Autopilot in Hardware-in-the-Loop (HITL) Simulation with UAV Dynamics in Simulink example.
Required Third-Party Software This example requires this third-party software:
Required Hardware To run this example, you will need the following hardware:
Workflow to run model on NVIDIA Jetson along with Pixhawk in HITL mode
The above diagram illustrates the PX4 and NVIDIA Jetson HITL setup and the physical communication between various modules.
This example uses four different Simulink models.
Simulink model for Flight Controller to be deployed on PX4 Autopilot.
Simulink model for Obstacle avoidance algorithm to be deployed on NVIDIA Jetson.
Simulink model for Autonomous algorithm to be deployed on NVIDIA Jetson.
Simulink model for flight visualization with Unreal Engine Simulation for Unmanned Aerial Vehicles.
To avoid performance degradation in MATLAB due to three different Simulink models running at the same time, launch three separate sessions of same MATLAB.
In the first session of MATLAB, the Flight Controller is deployed on Autopilot and the UAV Dynamics model will run on host computer communicating with Autopilot.
In the second session of MATLAB, the Simulink model for flight visualization with Unreal Engine Simulation will be running. This can be skipped if you opt to not add the flight visualization.
In the third session of MATLAB, the Simulink model for NVIDIA Jetson communicates with MATLAB on host computer using Monitor & Tune Simulation.
Step 1: Make Hardware Connections and setup the Pixhawk in HITL mode
1. Connect your Pixhawk board to the host computer using the USB cable.
2. Ensure that you have configured the Pixhawk board in HITL mode as documented in Setting Up PX4 Autopilot in Hardware-in-the-Loop (HITL) Mode from QGroundControl.
3. Ensure that you have setup the PX4 Firmware as mentioned in Set Up PX4 Firmware for Hardware-in-the-loop (HITL) Simulation.
4. Setup and configure your NVIDIA Jetson on network using MATLAB Coder Support Package for NVIDIA Jetson and NVIDIA DRIVE Platforms.
Step 2: Launch first session of MATLAB and the MATLAB Project
The support package includes an example MATLAB project having the PX4 flight controller and the UAV to follow the mission set in the QGroundControl (QGC).
1. Open MATLAB.
2. Open the example MATLAB project by executing this command at the MATLAB command prompt:
3. The Simulink project uses PX4demo_HITLSimulinkPlant as reference project. The UAV dynamics and autopilot controller are provided by the PX4demo_HITLSimulinkPlant project. Once the Simulink project is open, click the Project Shortcuts tab on the MATLAB window and click Open Autopilot Controller from the PX4demo_HITLSimulinkPlant drop-down list to launch PX4 Controller named Quadcopter_ControllerWithNavigation.
4. Navigate to Navigation subsystem. This is a Variant Subsystem with guidanceType as the variant control variable. Define guidanceType = 1 in the global workspace to choose the navigation subsystem for this example.
5. In the Project Shortcuts tab, from the PX4demo_HITLSimulinkPlant drop-down list, click Open UAV Dynamics to launch the Simulink UAV Dynamics model named UAV_Dynamics_Autopilot_Communication.
6. Open the Simulink Plant model UAV_Dynamics_Autopilot_Communication and configure the serial port. Select the serial port of Pixhawk which is connected to the host computer. Add the following UDP connections of onboard computer in the MAVLink Bridge blocks. For more information, see MAVLink Connectivity for QGC, On-board Computer and Simulink Plant. Double-click the MAVLink Bridge blocks to open the block Parameters dialog box.
a. Add the IP address of onboard computer (NVIDIA Jetson) in the MAVLink Bridge Source and MAVLink Bridge Sink blocks. Ensure that you can ping the NVIDIA Jetson successfully from the host PC. Enter the port number as 14540.
b. Add localhost connection for the flight visualization in the MAVLink Bridge Source block. Enter the port as 25000. Skip this if you are opting to not add the flight visualization.
7. Copy the MATLAB Project Path to clipboard.
Step 3: Configure Simulink Controller model for HITL mode
1. Follow the instructions mentioned in Configure Simulink Model for Deployment in Hardware-in-the-Loop (HITL) Simulation.
Note: These steps are not required in the pre-configured model. Perform these steps if you have changed the hardware or not using the pre-configured model.
2. Click Build, Deploy & Start from Deploy section of Hardware tab in the Simulink Toolstrip for the Controller model Quadcopter_ControllerWithNavigation.
The code will be generated for the Controller model and the same will be automatically deployed to the Pixhawk board (Pixhawk 4 in this example).
After the deployment is complete, QGroundControl will be automatically launched.
Note: If you are using Ubuntu, QGC might not launch automatically. To launch QGC, open Terminal and go to the location where QGC is downloaded and run the following command:
Step 4: Launch Second Session of MATLAB and Open the Flight Visualization Model
1. Open the second instance of the same MATLAB version. In this MATLAB session, the Simulink model for scenario simulation and flight visualization using unreal environment runs.
2. Navigate to the project path previously copied in Step 2 in the current MATLAB.
3. Click on the .prj file to launch the same Project in current MATLAB.
Ensure that your Host PC is Configured with MATLAB supported GPU (GPU with compute capability of more than 5).
4. In the Project Shortcuts tab, under the PX4demo_HITLSimulinkPlant dropdown, click Open 3D Visualization with Unreal Engine to launch the onboard model named Unreal_3DVisualization.
In this model, The MAVLink data from the PX4 Autopilot is received over UDP (port : 25000) and is used to decode the position and attitude data of the UAV. After coordinate conversion, it is then passed to the Simulation 3D UAV Vehicle block for flight visualization.
5. Enable the streaming of simulated depth sensor data NVIDIA Jetson by updating the variable
1 in MATLAB workspace.
6. The Simulation 3D Camera block provides the Camera image from the Unreal environment. In this example you stream the depth images from the camera block to NVIDIA Jetson using Video Send block. Double-click block to open the block Parameters dialog box. Add the IP address of onboard computer (NVIDIA Jetson) in the dialog box and click OK.
7. In this example, a Suburban scene is used for the Unreal simulation environment. To download this scene, run the following commands.
Verify the map is downloaded on your disk by running the following command:
8. Double click on the Simulation 3D scene Configuration block and select the Suburban scene from the dropdown for Scene name
9. On the Simulation tab, click Run to simulate the model. Once the model start running, you will see the Unreal simulation environment getting launched. A sample screen is shown below.
Step 5: Launch Third Session of MATLAB and the Onboard Model
1. Open the third instance of the same MATLAB version.
2. Navigate to the project path previously copied in Step 2 in current MATLAB.
3. Click on the .prj file to launch the same Project in current MATLAB.
4. In the Project Shortcuts tab, click Open Obstacle Avoidance Model to launch the onboard model named Onboard_ObstacleAvoidance.
This model implements the PX4 path planning interface using MAVLink Serializer and MAVLink Deserializer blocks. For more information, see PX4 Path Planning Interface. The MAVLink messages that are exchanged as part of this interface is shown in the following diagram.
5. This example demonstrates how to avoid obstacles by using the Obstacle Avoidance block from the UAV Toolbox.
6. Navigate to Target hardware resource > Board Parameters, enter the IP address of the NVIDIA Jetson and your login credentials.
7. Double-click the UDP Send block to open the block Parameters dialog box. Enter the IP address of the host PC on which you are running UAV_Dynamics_Autopilot_Communication as Remote IP address. Enter the value for Port as 14580. Ensure that the host computer and the onboard computer are connected to same network.
8. Configure the Network Video Receive block to receive the depth data from Unreal environment. Note that the port number and compression parameters in the Network Video Receive block are same as those of the corresponding Video Send block streaming camera image from Unreal engine.
9. Click Monitor & Tune from Run on Hardware section of Hardware tab in the Simulink Toolstrip.
The code will be generated for the Controller model and the same will be automatically deployed to NVIDIA Jetson. NVIDIA Jetson should start communicating with host over Monitor & Tune Simulation.
Note: Ensure that there are no other deployed models from Simulink are running in NVIDIA Jetson. If you are unable to verify, reboot the Pixhawk hardware board before starting the deployment.
The algorithm in NVIDIA Jetson also communicates with the Plant model UAV_Dynamics_Autopilot_Communication over UDP.
Step 6: Run the UAV Dynamics model, upload Mission from QGroundControl and fly the UAV
1. In the Simulink toolstrip of the Plant model (UAV_Dynamics_Autopilot_Communication), on the Simulation tab, click Run to simulate the model.
2. Set the PX4 parameter
COM_OBS_AVOID enabling the PX4 path planning interface. Navigate to Parameters from the main menu and set the
COM_OBS_AVOID parameter value to
3. Set the parameter
COM_DISARM_PRFLT value to
4. In the QGC, navigate to the Plan View.
5. This example provides a preplanned mission OA_simple_mission.plan that you can upload to QGC. To upload the preplanned mission, click Open Model button at the top of this page (MATLAB Help browser) to download the plan file (OA_simple_mission.plan). In the QGC, navigate and select OA_simple_mission.plan file.
After you upload the plan, the mission is visible in QGC.
6. Click Upload button in the QGC interface to upload the mission from QGroundControl.
7. Navigate to Fly View to view the uploaded mission.
8. Start the Mission in QGC. The UAV should follow the mission path.
9. Observe the flight visualization in the Unreal Engine. The drone deviates from the desired path to avoid the tree in the Unreal scene.
10. Go to the Onboard_ObstacleAvoidance model and validate the depth image streamed to Jetson during the flight in the Video Viewer (from Computer Vision Toolbox™).
While Simulating the visualization model in Step 4, you might get any STD exception errors such as, "some module could not be found".
Change the compiler to Microsoft Visual C++ 2019 using
mex --setup C++ command to fix the issue.
"Avoidance system not available" warning when starting a mission. This warning occurs because the communication between NVIDIA Jetson and PX4 HITL is not established.
Ensure that host PC and NVIDIA Jetson are connected to same network. Try pinging NVIDIA jetson from the host PC and vice versa. Also, double-check the NVIDIA Jetson IP address entered in the MAVLink Bridge blocks as mentioned in 6th point of Step 2 and Host PC IP address entered in the UDP send block as mentioned in 7th point of Step 5.