Acubed Surpasses Goal for Massive Data Collection Campaign
Acubed has reached a tremendous milestone: we have collected data from over 100 airports across the United States. This achievement is part of our Wayfinder team’s data collection campaign efforts to gather massive amounts of airport and runway imagery to mature machine learning solutions for a more automated era of aviation.
“We flew from Miami to Boston, San Diego to Seattle and all airports in between.”
Early collection efforts were mostly limited to the west coast, but Acubed announced the expansion of its data collection campaign to airports across the country in May 2023 at MIT’s annual AI conference, EmTech Digital. Paul Smith, Acubed’s director of flight test operations, explained that extending efforts to airports across the country gives Acubed access to a more diverse data pool, given each airport’s unique layout. To help identify which airports to visit, the team focused on those where the Airbus A320 - the most successful single-aisle aircraft in the world - operates.
Flight Test Lab
Pilots visited each airport in Acubed’s Flight Test Lab, a modified Beechcraft Baron 58. The experimental aircraft – equipped with cameras and sensors – allows the team to explore systems that can be applied to a variety of commercial aircraft, including the A320. “One of the reasons we picked the Beechcraft Baron,” Smith said, “is it’s very similar to airliners in the way it performs in low-altitude regions.” Systems developed using Acubed’s Flight Test Lab have already contributed to world firsts in autonomous flight: taxiing, taking off and landing of commercial aircraft, as well as performing emergency operations to cruise and land in the unlikely event of crew incapacitation.
Smith and two other pilots flew the data collection operations. In the northeast, where airports are located more closely together, pilots were able to visit as many as 20 per week. In the west, where they are more spread out, resulting in longer flight times, airport visits averaged 10 per week. Although pilots flew solo, the project has been a team effort, from briefings and coordinating with air traffic control, to analyzing collected data and exploring corner cases in simulation.
Gathering imagery from an array of airports has enabled Acubed researchers to mature machine learning (ML) algorithms by creating a more robust data pool used to train and test systems. It also gives the team a larger template to which they can apply synthetic datasets. “To go collect the data to the level that we need for each airport would be impossible,” Smith said.
Synthetic datasets allow researchers to better train and test ML algorithms. Using synthetic data, the team can take imagery collected from airports and supplement it with synthetic data to explore a variety of conditions and scenarios in simulation, from obstacles on the runway to varied flight conditions.
The campaign has resulted in fortuitous industry collaborations as well. In part, it has led to a Memorandum of Understanding with Dallas / Fort Worth International Airport (DFW) to develop potential solutions to increase the safety, efficiency and sustainability of airport ground operations. The collaboration will explore how Acubed’s ML and AI systems can support some of the challenges DFW, one of the world’s busiest airports, faces.
Smith explained that the century mark also helps to validate the data collection concept. “We have two challenges,” Smith said. “One is to build the synthetic data and the other is to prove that the synthetic data we build will replicate real life in the models we are running. So what we’re doing is taking the real data that we collect in the airplane and comparing it to the synthetic data in enough cases for a regulator to find acceptable.”
The approach has worked. The research resulting from the collected data has validated Wayfinder’s method, helping the team reach Technology Readiness Level (TRL) 3 and establish proof-of-concept as they progress toward TRL 4. Wayfinder will continue its data collection campaign, visiting airports and supplementing real data with synthetic, in order to mature the program further. “We can never have too much data,” Smith said. “Every flight and, indeed, every approach adds different details to the picture!”
Quality data is integral to ML research. Reaching, and surpassing this 100-airport milestone represents a critical step in developing a robust data pool. The milestone, and the expert research it enables, further Acubed’s leading role in maturing ML solutions and promoting the next generation of flight. We look forward to increased data collection across a variety of weather and lighting conditions in 2024.