This article is published in Aviation Daily part of Aviation Week Intelligence Network (AWIN), and is complimentary through Nov 26, 2024. For information on becoming an AWIN Member to access more content like this, click here.
PARIS—Thales is investigating the use of a neural network combined with a certified conflict detection system to help prevent aircraft from flying too close to one another in a given airspace.
The emergence of such a hybrid artificial intelligence (AI) system in air traffic management (ATM) could relieve pressure on air traffic controllers, who would ultimately decide whether to use a proposed resolution advisory. It could also improve avoidance trajectories, for instance, by making them shorter. Such a system might enter service around 2030, after more development and certification work is conducted.
Under the conventional procedure when a trajectory conflict—or potential loss of separation—is identified, the air traffic controller steps in and identifies a new trajectory. They provide the crew with an instruction, or clearance, to alter their flightpath and avoid the conflict. Simultaneously, controllers make sure they avoid conflicts with other aircraft in the same airspace.
Regulations call for maintaining separations of 5 nm horizontally and 1,000 ft. vertically—some variations may apply. “When traffic is very dense, this can be quite stressful,” Thales Airspace Mobility Solutions chief data scientist Andrei Purica said at the AI Applications In Aeronautics, Defense and Space conference organized by the Air and Space Academy here Nov. 13-14.
Thales already offers an automated resolver. Based on classical research algorithms, it works in the background, looking for an alternative trajectory, Purica said. The resolver shares the result with the controller, who may decide to use it or do something else.
The current model has limitations, however. Due to computational power limits, it cannot analyze every single solution in the background. It may thus miss some, Purica pointed out.
Thales is thus exploring AI and, more specifically, reinforced learning. The approach gives more exhaustive and faster results and includes cost functions—the benefits and drawbacks of each solution. It can take into account criteria such as CO2 emissions and contrail reduction, Purica adds.
Among the limitations of AI for ATM, safety concerns rank first. The system cannot provide a solution guarantee, and a solution may be difficult to explain. “In accuracy, 99% is not enough,” Purica said.
Hence Thales’ idea for a hybrid approach, combining AI and symbolic calculations. “If a conflict is detected, we ask the neural network what he thinks of a certain option with respect to a certain cost function,” Purica said. “We look at rewards and penalties. In a flightpath’s deviation, the neural network may suggest a range of angles is better.” The classical tool then focuses the search and gives the controller a validated solution. To train the AI model, Thales relied on 20-30 min. ATM scenarios and ensured it was learning from failures, Purica said. The experiment was deemed successful. “We ran the system for several days,” he said. “After 15 million clearances, the system began to understand and started to maximize the cost function we provided. After 20 million clearances, it not only solved 99% of the problems, it reduced the average length of the trajectories.”
In real life, the medium-term conflict detection system—an existing certified module—would validate the AI tool’s suggestion. Purica anticipates it may take 5-7 years before such a system is put into operation. Thales tested the AI part of the system with controllers, who found it acceptable—sometimes acting like a novice controller, sometimes proficient. The company’s engineers hope to experiment with the hybrid approach in the near future, Purica said.
