A group of researchers at Carnegie Mellon University (CMU) has developed what they call the first AI pilot that enables autonomous aircraft to navigate a crowded airspace.
The artificial intelligence (AI) system safely avoids collisions, predicts the intent of other aircraft, and coordinates with their actions. It can also communicate with pilots and air traffic controllers over the radio. The team says they want to develop the AI so that it is eventually indistinguishable from a human pilot.
Passing the Turing Test
Jean Oh is an associate research professor at CMU’s Robotics Institute (RI) and a member of the AI pilot team.
“We believe we could eventually pass the Turing Test,” Oh said.
The AI relies on vision and natural language to communicate with other aircraft similarly to a human pilot, which helps achieve safe and socially compliant navigation. The team achieved this coordination by training the system on data collected at the Allegheny County Airport and the Pittsburgh-Butler Regional Airport. The data included air traffic patterns, images of aircraft, and radio transmissions.
The AI is equipped with six cameras and a computer vision system that helps it detect nearby aircraft. It also has an automatic speech recognition function that uses natural language processing (NLP) to understand incoming radio messages and communicate with pilots and air traffic controllers through speech.
Navigating Crowded Airspace
Because the current airspace is already crowded, the FAA and NASA have proposed dividing urban airspace into lanes or corridors that have restrictions on when, what type, and how many aircraft can use them. This proposal would result in a big change for current use and standard practices and could result in problems like air traffic jams.
There are many challenges involving the development of an AI to operate in the crowded and pilot-controlled lower-altitude traffic under visual flight rules (VFR), but the team’s system is designed to seamlessly interact with aircraft in this environment.
Sebastian Scherer is an associate research professor in the RI and a member of the team.
“This is the first AI pilot that works in the current airspace,” said Scherer. “I don’t see that airspace changing for UAVs. The UAVs will have to change for the airspace.”
The AI pilot has not yet been tested on an actual aircraft, but it has performed impressively with flight simulations, where it demonstrated an ability to navigate around other piloted aircraft.
Jay Patrikar is a Ph.D. student in the RI. He was involved with the project.
“We need more pilots, and AI can help,” Patrikar said.