Star Wars: Coming soon to the skies near you
Anyone who has flown has experienced disruptions and delays. The National Airspace System (NAS), overseen by the FAA, is working with academia and industry to move to a new system called Next Generation Air Transportation System (NextGen). The system will include new technologies and procedures to increase safety, efficiency, capacity, access, flexibility, predictability and resilience, while reducing the environmental impact of aviation.
Our national airspace, already a complex system, is only going to get more complicated with the proliferation of unmanned autonomous aerial vehicles (UAVs). There are uncertainties in almost every aspect of the airspace. Factors include environmental constraints (like wind and precipitation); scheduling (departures and arrivals); and vehicle dynamics, which cannot be mathematically described with 100 percent accuracy.
All these restrictions affect the safety and efficiency of operations. For example, if a flight is delayed due to weather conditions, air traffic controllers may have to reschedule other flights, which usually generates delays. Current operations are based mostly on playbooks, which generally are very conservative. Some software decision support tools have been used to ease the work of air traffic controllers, but the applications are limited. As a result, there is very little autonomy in our NAS today.
Colleagues and I are seeking to change that situation. My research focuses on machine learning and artificial intelligence (AI) for unmanned aerial vehicles, which present a major change and disruption to our traditional aviation system. I am working on machine learning algorithms for visual object detection and recognition in real time. This is especially challenging due to the imaging conditions and constrained onboard computing resources on UAVs, which usually are very small and have a meager payload and battery/power resources.
Machine learning needs high-end graphics processing units (GPUs), which consume too much payload and power for use on UAVs. While the deep learning models could be trained at ground stations, UAVs are constantly collecting new data from changing conditions and must keep training these models for real-time decision support.
I want to efficiently train deep-learning models for lightweight and low-cost aerial vehicles onboard the UAVs themselves. To the best of my knowledge, Purdue’s School of Aeronautics and Astronautics is among the first academic units to investigate this approach. Professors Inseok Hwang and James Goppert and I are validating our machine algorithms and testing them in a newly-built indoor UAV test center at Purdue University Airport. We also are collaborating with industrial partners, as well as Professors Xiaoqian “Joy” Wang in the Purdue School of Electrical and Computer Engineering and Heng Huang from the University of Pittsburgh, both leading experts in machine learning and AI.
In the movie Star Wars: The Rise of Skywalker, swarms of aerial vehicles communicate with their neighbors and make collaborative decisions to achieve their individual and global missions. This scenario is very similar to our future aviation system in several ways: Each vehicle (crewed or uncrewed) needs to arrive at a destination; deliver people or goods at a scheduled time; avert collisions with other planes; avoid dangerous zones (such as areas with severe weather); and respect its own dynamics/performance constraints.
There will be many more aircraft in our skies in the near future, many of them uncrewed. It is critical for each vehicle to be able to sense its neighbors and environment; compute to make decisions (like “learning” neighbors’ behaviors and the environment’s impact); and communicate with neighbors, including ground stations. Some of these actions will be fully autonomous, while others will be manual, taken by pilots and air traffic controllers.
The increasingly crowded and complex networked Internet of things (IoT) in the sky needs a better understanding of system dynamics, along with better models and algorithms, to make the aviation system safe yet still efficient. Our Purdue Engineering efforts are fueling progress in that direction.
Dengfeng Sun, PhD
Associate Professor
School of Aeronautics and Astronautics
College of Engineering, Purdue University
Related Links
Purdue UAS research and test facility, Purdue University Airport
Purdue Engineering Review: “Urban Air Mobility: Welcome to the ‘Third Era of Aviation’”
