Speech recognition system can open locks for drone delivery

It seems inevitable at this point. Our urban skies quickly fill with all kinds of multicopter aircraft, from small logistics and delivery drones to eVTOL air taxis and minibuses. This brings many positives, such as speed, convenience, reduced emissions, and driving out of the road. There are some downsides, especially noise pollution, the risk of things falling into your head, and the evaporation of certain senses of privacy. Constant eyes in the sky.

In any case, this future is coming, and it’s only the office work that really hinders things right now. One of the key issues is safety concerns. How can hundreds or thousands of these small aircraft share airspace without colliding with each other or interfering with the flight path of large aircraft? This issue needs to be comprehensively resolved before aviation authorities can open the door to a drone that operates “beyond the line of sight” of a remote pilot.

Zipline, an autonomous drone delivery company, has already completed more than 300,000 commercial deliveries in the controlled test market over the past five years, providing an attractive solution in the form of an acoustic detection avoidance (DAA) system. I think it is. Founder and CTO Keenan Wyrobek describes it as the “Holy Grail of Drone Technology.”

Each aircraft is equipped with a microphone array and on-board processor to detect other aircraft up to 2 km (1.24 miles) away, provide 360-degree visibility, and a flight control system monitors and tracks multiple aircraft. I will be able to do it. We will adjust the appropriate flight route on the spot.

The acoustic detection and avoidance system provides autonomous aircraft with a spherical recognition zone with a maximum radius of 2 km.

The acoustic detection and avoidance system provides autonomous aircraft with a spherical recognition zone with a maximum radius of 2 km.

Zipline International

There are many other solutions, including optical detection systems like Iris Automation’s Casia360 system, multimode and radar-based systems that General Atomics has tested on predator drones for many years, and other transponders. A style idea where the aircraft should always broadcast its position.

Zipline’s acoustic solution seems to offer some significant benefits. Radars and camera arrays are bulky, heavy, expensive, and provide no awareness behind visual obstacles such as hills and buildings. The camera has a hard time finding other drones in the distance and can be endangered by lighting conditions and weather. Transponders rely on terrestrial communications and the assumption that everyone else is playing according to the rules.

Microphone arrays, on the other hand, are small, lightweight, self-contained, can provide spherical coverage, have the potential to capture information from behind specific obstacles, and are about as cheap as sensor devices. Of course, acoustic systems have their own challenges. The drone to which they are connected produces itself multiple highly variable audio sources. These audio sources need to be deducted very accurately from the feed of each mic if they remain sensitive to small propellers that are more than a mile away.

Wind noise is another issue, especially in speed. For example, motorcycle intercom systems between riders have been impressed with wind noise reduction technology in the past, but there is a big difference in the delivery of easy-to-understand audio. It sends a signal between two people and provides a very clean spherical audio feed that can accurately analyze the small audio signals needed for FAA grade sense and evasion.

The DAA system uses a time-tuned beam formation feed from an array of microphones to run a deep learning model to estimate the probability that a particular sound signature is coming from another aircraft and track its movement. increase

The DAA system uses a time-tuned beam formation feed from an array of microphones to run a deep learning model to estimate the probability that a particular sound signature is coming from another aircraft and track its movement. increase

Zipline International

In a US patent application, Zipline could allow a DAA-equipped drone to intentionally turn around to better read the object it’s tracking, and as a way to filter out the shape of the drone between microphones. Describes a system that uses variability. Unique noise signal. It uses beamforming with time-tuned audio signals from each microphone and passes the resulting signal through a deep learning model to estimate whether a particular signal was produced by the aircraft.

“We have created a system that is agile enough to work with the highest margins, and we have created a system that we can think of and adjust in real time. DAA has many years of development and number. The result of 100,000 hours of flight time. This is an elegant answer to the challenge of flying beyond the line of sight of the United States. Today, not ten years later. “

Zipline founder and CEO Keller Rinaudo said, “DAA is the cornerstone of expanding immediate delivery in the United States and around the world. This system has become the industry standard for safe flight of all commercial autonomous aircraft. I’m imagining a future that will be. ”

According to the company, the DAA system is already built into its drones and seeks regulatory approval before it can be activated in various regions.

The time it takes to approve it and the exact format it will take have not yet been announced. According to Zipline, DAA systems can integrate well with other systems as needed. I am very interested in seeing how it has been tested and how it works with different targets in a complex and crowded audio environment. However, it certainly seems like a promising development, and it is possible to add spherical recognition capabilities to UAVs without many major drawbacks. And if it does what is written on the can, it could be an important step towards tomorrow’s bustling sky.

Source: Zipline

Leave a Comment