Ready for takeoff? This guide outlines key steps to review before building and flying your autonomous drone. 

Step 1: Choose Your Platform

Your first step is selecting the right hardware platform for your project. ModalAI offers a range of options to suit various needs and skill levels:

• Development Drones: For developers wanting a quick start, ModalAI's pre-built drones, like the Starling 2 or Starling 2 Max, provide a fully-integrated hardware and software platform for rapid prototyping and development. These drones are NDAA-compliant, making them an ideal choice for government and enterprise clients.
• Autopilots: Developers who want maximum control over the entire design process can assemble a drone from ModalAI’s VOXL autopilots, building from the ground up. This approach offers unparalleled flexibility but requires more technical expertise. Using VOXL 2 or VOXL 2 Mini autopilots can significantly boost development time. ModalAI also offers a comprehensive suite of drone accessories, ranging from image sensors, radios, and accessories, so you can create your perfect drone. 

Step 2: Hardware Setup and Validation

This step focuses on ensuring all hardware is correctly assembled and functioning. Here, you will double check that your hardware is set up correctly.

Validation steps might include:

• Airframe Construction: Ensure the drone's airframe is robust and structurally sound.
• Companion Computer Calibration: Connect your companion computer to its ground control station and flight controller. Configure network connection between drone and ground control station via one of the supported methods (WiFi, 5G, Microhard, Doodle Labs, etc.) in order to view live data from the drone.
• Motor and Propeller Installation: Correctly mount and secure motors and propellers, ensuring they are spinning in the right direction and appropriately balanced.
• ESC Calibration: Calibrate electronic speed controllers for proper motor control.
• Sensor Integration: Integrate and test any additional sensors, like GPS modules, IMUs, and cameras.
• Voltage Testing: Check for expected voltage levels and validate that the current consumption of the vehicle is expected. 

Step 3: Software Setup and Configuration

ModalAI's VOXL SDK provides a user-friendly development environment to program and configure your drone. The regularly updated VOXL SDK provides streamlined configuration, faster video processing, and improved robustness. Key features include:

• Pre-programmed Autonomous Navigation: VOXL SDK includes pre-programmed algorithms for Visual Inertial Odometry (VIO), Visual Obstacle Avoidance (VOA), and 3D mapping, significantly reducing development time.
• Integration with Popular Platforms: The SDK seamlessly integrates with PX4, ROS 2, OpenCV, and TensorFlow Lite, providing broad compatibility with existing development tools and workflows.
• Open Documentation: VOXL SDK is open, meaning that developers can access our code, software set up, and tutorials free of charge. 

Step 4: First Flight Set Up

Before your first flight, you must configure your drone’s start up behavior. In this step, you’ll determine which features and software should be activated when your drone is turned on. This process ensures the accurate functioning and communication of your drone hardware to your software. Key considerations include:

• Linux Services: Determine which Linux services will run and in what order.
• Modem Configuration: Ensure proper modem (Microhard, LTE, 5G, Doodle Labs) connection to your drone. 
• Video Streaming: Set up video streaming and ensure all image sensors are working.
• Docker: Configure any Docker containers to run.

Step 5: Autonomous Flight and Mission Planning

After completing the preceding steps, you can start testing your drone’s autonomous capabilities. Begin with simple tests and gradually increase complexity. Remember to always have a pilot ready to manually take control if necessary. The "How to Fly Your Drone Like a Pro" blog offers helpful guidance on flight techniques, both manual and autonomous. 

Building an autonomous drone can seem daunting, but ModalAI provides comprehensive tools, resources, and support to streamline the process. For further support, visit ModalAI's website, access the comprehensive documentation, or participate in the ModalAI forum.

A look at the top 5 companion computers used by drone developers.

A companion computer is a separate onboard computer that works alongside the UAV’s primary flight controller. The companion computer acts as the brains of the robot, and processes image sensor and video data for “sight” and adds context to it - enabling perception. Drones that have a companion computer onboard can process video and image data in real time and execute autonomous navigation, path planning, and Computer Vision (CV)/AI-tasks such as obstacle avoidance and object detection. Add a radio to the mix and drones can operate and stream encoded video beyond the workshop walls. Ultimately, companion computers enable developers to run advanced AI and CV applications unique to their drone use case.

In the use case of the drone, there’s another critical sub-system in play: the flight controller. The flight controller communicates with the companion computer, adding context to data like speed, direction, and altitude, also known as PVAT (Position, Velocity, Acceleration, and Time). Traditionally, a flight controller has been a separate add-on unit, but if you can find one pre-configured with a companion computer, that can be a more efficient option.

ModalAI VOXL 2

VOXL 2 is a Blue UAS Framework autopilot that fuses a PX4 or ArduPilot flight controller and companion computer powered by Qualcomm QRB5165 into one 16g system-on-module (SOM). This credit-card-sized autopilot is built in the USA and is NDAA ‘20 Section 848 compliant. VOXL 2 features 15 TOPS of AI embedded NPU and can conduct autonomous computer vision based navigation and stream real-time h.265 video over 4G/5G, WiFi, Microhard and Doodle Labs networks. Unique to the VOXL 2 is a dedicated computer vision DSP, which offloads high-power image processing, freeing up power consumption from the GPU and CPU. VOXL 2-powered drones can conduct real-time object detection, tracking, and classification, as well as other machine learning tasks out of the box. 

VOXL 2 is used by drone OEMs in the ISR, agriculture, and defense space, researchers, and individual developers. Developers can get started quickly with VOXL 2’s open documentation and compatibility with popular developer applications such as PX4, ArduPilot, Docker, OpenCV, OpenCL and TensorFlow Lite.

VOXL 2 starts at $1,249 and is available for purchase through ModalAI’s website.

NVIDIA Jetson TX2 Developer Kit

The Jetson TX2 is a 7.5-watt supercomputer on a module (SOM) that delivers 1.33 TFLOPs, NVIDIA Pascal GPU with 256 CUDA cores and ARM Cortex-A57 CPU. The Jetson TX2 is designed for embedded systems and AI applications and works with NVIDIA’s rich set of AI tools and workflows, which enable developers to train and deploy neural networks quickly. The Jetson TX2 is supported by the NVIDIA JetPack SDK and supports popular AI frameworks like TensorFlow, PyTorch and Caffe. 

Although the Jetson TX2 boasts powerful performance, the 314g SOM alone cannot power an autonomous drone. Developers who choose to use the Jetson TX2 for autonomous drone applications need to configure the Jetson TX2 with a developer kit to access additional IO ports that will enable communication to flight hardware and a separate flight controller to activate flight. These additional boards can add significant weight to a drone’s final design and should be taken into consideration during the evaluation phase.  

The Jetson TX2 developer kit has recently been discontinued. The Jetson TX2 is $490 and is available for purchase through Arrow’s website.

NVIDIA Orin Nano Developer Kit

The Orin Nano is an entry-level companion computer that delivers up to 40 TOPS of AI performance in the smallest Jetson form factor. The Orin Nano features NVIDIA Ampere GPU with 1024 CUDA cores and 32 Tensor Cores for running complex AI models and deep learning algorithms. Similar to the Jetson TX2, the Orin Nano is supported by the NVIDIA JetPack SDK and supports TensorFlow, PyTorch and Caffe. 

Similar to the Jetson TX2, the 140g Orin Nano requires a separate flight controller to activate flight. These additional boards can add significant weight to a drone’s final design and should be taken into consideration during the evaluation phase. 

The Orin Nano Developer Kit starts at $499 and is available for purchase at Arrow’s website.

Raspberry Pi 5

Raspberry Pi is a popular intro-level companion computer. At under $100, the Raspberry Pi is a great option for hobbyists, university researchers, and developers who want to get started on their low-compute project quickly at a low cost. Raspberry Pi is powered by an ARM-based processor and primarily runs on Raspberry Pi OS. Its affordability, coupled with powerful hardware and a rich ecosystem, makes it an excellent choice for both hobbyists and professional developers in the drone industry.

Some nuances with the Raspberry Pi developers should take note of is the low power GPU and lack of neural processing unit (NPU) and on-board flight controller. Developers looking to use a Raspberry Pi for autonomous AI drone applications will need to purchase a separate flight controller in order for the drone to intelligently process its surroundings. The Raspberry Pi’s lower power GPU makes it difficult to offload intense or sophisticated programming. Popular applications like PyTorch or TensorFlow Lite may experience latency because of this. 

Raspberry Pi 5 starts at $79 and is available for purchase from many approved resellers. 

Raspberry Pi 4

Raspberry Pi 4 is a versatile, low-cost, credit-card-sized single-board computer that is designed to promote computer science education and provide a powerful platform for various projects and applications. The open-source Raspberry Pi platform enables developers to create custom machine learning applications tailored to specific drone missions.  

Similar to the Raspberry Pi 5, Raspberry Pi 4 does not have a neural processing unit (NPU) and on-board flight controller. Developers looking to use a Raspberry Pi for autonomous AI drone applications will need to purchase a separate flight controller in order for the drone to intelligently process its surroundings. 

Raspberry Pi 4 starts at $35 and is available for purchase from many approved resellers. 

At a Glance Comparison: Companion Computer Developer Kits

The following table includes stats from SOM alone and separate developer kits

*Source: https://forums.developer.nvidia.com/t/weight-of-boards/175292/3 

SDK 1.1.2 is now available on all VOXL products.

Happy New Year! We’ve been hard at work making big changes under the hood. Our latest VOXL SDK 1.1.2 is now available on all VOXL products - including our original Blue UAS Framework 1.0 VOXL line. Experience streamlined configuration and faster video processing like never before.

SDK 1.1.2 feature highlights:

APQ8096 and Qualcomm Flight RB5 5G Platform support

Now all VOXL and RB5 products run the same SDK across the board

v1.14 on Flight Core v1 and v2
Our Flight Core V1 and V2 images are now built from the same codebase as the PX4 version running on VOXL 2 SDSP ensuring predictable PX4 behavior across the product line.

Hardware-accelerated video compression in voxl-camera-server

Always-on RTSP video and MAVLink camera protocol support

Streamlined configuration

Other

• Improved thresholding in voxl-calibrate-camera that’s less sensitive to lighting conditions
• 16-bit raw lepton thermal camera data pipe
• Send VIO packets with quality=-1 to PX4 on VIO failure to utilize PX4’s new quality metric functionality and make for more graceful fallback to altitude flight mode.
• Robustness improvements to libmodal_pipe