Compiled by the editorial staff of eVTOL.com


More progress for Airbus’s Wayfinder autonomy project

Airbus’s Silicon Valley Innovation Center, A3, has released a new video highlighting the progress of its Wayfinder project, which is developing scalable, certifiable autonomy systems for self-piloted aircraft.

Airbus Wayfinder simulated approach
Initially developed for the Vahana eVTOL demonstrator, Wayfinder is now being used in other Airbus projects, including the Autonomous Taxi, Take-Off and Landing (ATTOL) demonstrator. A3 by Airbus Image

Launched in 2016 to develop the sense-and-avoid system for the Vahana eVTOL demonstrator, Wayfinder has now been extended to vehicle applications across the Airbus product line. That includes Airbus’s Autonomous Taxi, Take-Off and Landing (ATTOL) demonstrator, an A320 that will be equipped with “a suite of sensors, actuators, and computers to explore the potential of autonomy via computer vision and machine learning,” the Wayfinder team wrote on its blog.

The new video shows a Wayfinder machine learning algorithm in action during a simulated approach to a runway at Toulouse-Blagnac airport. This type of simulation allows the team to quickly develop and test different versions of its software as it works to “teach the aircraft to see and navigate to the runway using deep learning,” as Wayfinder head of software Harvest Zhang explained in a July 10 blog post.

According to Zhang, Wayfinder used the X-Plane Pro flight simulator to generate photorealistic synthetic data for training the convolutional neural network that forms the basis of its computer vision system for ATTOL. “Using synthetic data, our network can reliably detect the runway from several miles away, and runway distance estimates are typically within a few percent of ground truth,” he wrote, acknowledging that “localizer and glideslope deviation estimates are more challenging and need some more work.”

Zhang said that the network performed “surprising well” on real images, despite having only been trained on synthetic images. “As we collect more real flight data, we can begin to explore domain adaptation across different combinations of synthetic and real training data, while recent developments in generative adversarial networks may allow us to blur the distinction between synthetic and real data altogether,” he added.

As the video describes, for Vahana, the Wayfinder team captured training data through the use of drones that emulated how and where the vehicle would fly, and the obstacles it would encounter. The full-scale demonstrator Vahana Alpha One has now completed over 70 test flights with the Wayfinder system on board.

Leave a comment

Your email address will not be published.