Qualcomm Brings ML and 5G Communication to Drones
Communication is often a pain point in drone design. Now, Qualcomm says it is leveraging "swarm support" and its tensor accelerator to vault these challenges.
Back in March, Qualcomm made headlines when the Qualcomm Flight Platform debuted in NASA's Ingenuity helicopter for Mars missions.
During this mission, the developers found that flight challenges in space are even more pressing than those on earth since signals took minutes to arrive. Qualcomm and NASA researchers had to design for the unmanned aerial vehicle (UAV) to receive commands that were transmitted millions of miles away with a delay of up to 22 minutes.
Another challenge was ensuring the helicopter system consumed low power while computing high amounts of data. Qualcomm’s Flight Platform had a unique advantage over conventional drone technology because of its image signal processors and neural processing units.
The Qualcomm Flight Pattern powered the Ingenuity helicopter. Image used courtesy of Qualcomm
Now, Qualcomm has upgraded this platform—the Flight RB5 5G platform—to address communication challenges found in aerial inspections, first response rescue, and both commercial and industrial applications.
The Challenges of Current UAV Communication
An autonomous drone can reach difficult access points in various industrial, chemical, and marine environments. They can even provide rescue and relief in emergency situations. Given these complicated flight paths, UAV communication is a major challenge for drone developers.
Communication technology in drones. Image used courtesy of the LNM Institute of Information and Technology (PDF)
Whether on solo flights or in fleets, UAVs need various types of wireless channels and network protocols to establish a secure, reliable, and robust connection. To avoid breaks in incoming signals, communication networks need a:
- Line of sight
- Point-to-point link between each drone
- Fixed device such as a transmitter/receiver
Adding Machine Learning to Drones
As Qualcomm demonstrated, next-gen drone designs often integrate AI and a cellular connection to improve communication.
In a study from the LNM Institute of Information and Technology in India, researchers found that machine learning and path planning can be integrated into drones to improve network communication (PDF). That said, adding AI to drones also runs the risk of possible data leakage from system security. When drone cameras capture images, the device will need secure encryption to avoid consuming too much power away from the processing unit.
Cellular-connected UAV networks are another possible design option developers can use to establish stronger UAV communication; however, they come with more challenges such as frequency disturbances, rate adaptation, high altitude performance, and high power consumption.
Qualcomm Brings 5G and AI to Drone Flight
Qualcomm says its Flight RB5 5G platform is the industry’s first 5G and AI drone solution, complete with a reference design to give manufacturers plenty of design flexibility and an extensive range of connectivity for UAV pilots.
Included in the RB5 platform is Qualcomm's Spectra 480 image signal processor, a camera capable of processing 2 Gigapixels per second, capturing 200-megapixel photos, and delivering 8K video recording. By utilizing this platform, a UAV can have seven concurrent cameras at 15 tera operations per second (TOPS) processing system speed.
The Qualcomm RB5 5G Flight Platform is said to offer heterogeneous computing. Screenshot used courtesy of Qualcomm
Because drones struggle with communication when operating beyond a visual line of sight (especially in remote locations), the RB5 5G platform is equipped with something called "swarm support,"—a unique attribute that allows a fleet of drones to simultaneously carry out individual or collective tasks without requiring line of sight networks.
Real-time data is handled, stored, and delivered through the system’s LPDDR5, which has a memory speed of 2133 MHz and a CPU clock speed of 2.84 GHz. Qualcomm’s technology is built on a 7nm process. AI and machine learning come into play with RB5’s dedicated tensor accelerator.
Qualcomm’s Neural Processing SDK is said to optimize deep learning as a resource for secure edge computing and audio, imaging, embedded vision to establish path planning methods for the UAV.
Have you ever worked on drone communication designs? How did you overcome line-of-sight issues? Share your experiences in the comments below.