Marc Piette is founder and CEO of Xwing, a California-based startup developing technologies for autonomous aircraft. In this Q&A he discusses how the development of autonomous aircraft can be commercially focused and why fully autonomous aircraft could be in use before fully autonomous cars.
What has the company been doing since the funding was announced in August?
We’ve entered a partnership with Bell and we’ve been selected by NASA to participate in their “UAS in the NAS” program, which aims to tackle the key challenges that prevent commercial operations of unmanned aircraft in the US national airspace. It’s a two-year program covering the two key technologies – detect and avoid (DAA) technology and command and control (C2).
We’ve teamed up with Bell, Textron Systems and the University of Massachusetts’ CASA. Bell is providing their APT Cargo UAV platform, Textron will provide the C2 link and Xwing is providing the DAA system.
What’s the focus of the company’s R&D?
Our near-term focus is on developing a market leading DAA system. It will be able to handle multiple concurrent threats to navigate the airspace safely. It’s a key component on the roadmap to fully autonomous flight.
Is detect and avoid the biggest problem for autonomous aircraft?
It is one of the first problems that needs to be tackled. We don’t think of autonomy for aircraft as a binary thing, where you go from manually piloted to fully autonomous vehicles. Autonomous functions will be gradually developed, certified and integrated into aircraft over the next few years until you finally get a fully autonomous aircraft.
The key driver for this pathway is operational cost reduction. For single-pilot aircraft, the first significant step to reduce costs is to remove the pilot from the aircraft itself and put an operator on the ground, to do things like interfacing with air traffic control or validate decisions made by the aircraft.
At first ground-based operation of these aircraft will be performed on a one-to-one basis. It’s worth stressing these people will be operators, not pilots. They won’t have full control over the aircraft. But as we start reducing their responsibilities, by automating path planning, collision avoidance or landings for example, we will be able to move to a ‘many aircraft to one operator’ model and reduce costs.
Will people be comfortable with pilotless planes?
Some companies will be testing that assumption soon. There are a few companies that are looking to fly passengers without pilots on board within the next few years. They will need certifiable DAA functionality and C2 links to do that commercially in the US and Europe. In general, as is almost always the case, there will be early adopters. Those people will spread the word and encourage others to try and the market will expand from there. If people see significant benefits to adoption, they will overcome their initial apprehension. I believe a significant portion of the population will embrace it faster than most think.
How does the development of autonomous systems for aircraft compare to the development of autonomous cars?
Some pieces are harder, some are simpler. With cars you have to deal with a more diverse environment. That requires sophisticated AI techniques that haven’t yet matured.
Aviation deals with a much more controlled environment, which is beneficial. On the other hand, the safety levels required and the added challenge of bringing the aircraft to a safe stop when components do fail, make it a challenging problem. You need very high levels of reliability in aircraft as failure of some key systems can quickly lead to loss of life.
Level 5 autonomy for cars on most or all roads, where the steering wheel and pedals have been removed, might even happen after aircraft become fully autonomous. That’s not to say that you won’t see car operate commercially autonomously on some roads. Some companies like Waymo are leading the pack, but for a car to be able to operate in all environments fully autonomously will take a while.
It’s interesting that the transition phase of remotely operated aircraft is something autonomous car companies are contemplating. For example, the company Cruise is, I believe, considering having remote operators take over cars when they are stuck or in certain situations.
When will you have an aircraft flight testing your autonomous systems?
You may have seen a few projects that were called autonomous, but which were really highly automated. Projects like Aurora’s Centaur, a DA42 retrofit, the AACUS huey retrofit, , Rockwell/Leonardo’s SW-4 ‘solo’ or Scion UAS’s SA-400 Jackal among others are highly automated – but not yet autonomous. They need a more comprehensive perception stack to be fully aware of their surroundings, they are carrying out preset flight plans with advanced control systems and autopilots. It does looks impressive, but those vehicles lack awareness of their surroundings and the ability to make decisions on their own.
We could produce a similar technology demonstrator in a reasonably short time frame. But that’s not our core focus. We’re focused on getting the right route market to advance the industry significantly – the first step on that journey is to fly passengers or cargo commercially at lower costs and to alleviate the commercial pilot shortage issue.
That means supporting the commercial operation of pilotless aircraft, with ground operators through certifiable DAA and C2 technologies.
What’s your thoughts on how we should test autonomous aircraft?
There’s still a lot of open questions here. We’re talking about complex systems and algorithms that can’t realistically be shown to be safe by means of formal proofs. Another challenge is where we don’t always have a lot of data on physical component failure rates.
There are different approaches we can take to ensure system safety. We have access to tools that we didn’t have 10-15 years ago and can run very large-scale realistic simulations in the cloud. In a black box approach, we are considering the use of Monte Carlo simulations to simulate millions of flight hours with probability distributions of the inputs that the system or any of its components are likely to face. Some of those distributions should reflect the likely environmental conditions and encounters with other types of aircrafts. You can also have automated verification of outputs to ensure the values are bounded and within norms.
Beyond this black box approach, simulating failure scenarios and ensuring the system continues to operate safely is key and something the industry has already practiced. Aerospace has been making great strides in safety over the past decades so it’s no surprise that we’ve already come a long way in our approach.
An interim step of hardware-in-the-loop simulations (HILSIM) is also done to validate hardware integration.
Some of those methods will go a long way to give the FAA the levels of confidence that make them comfortable here. This is the ultimate goal – to demonstrate the safety of these systems.
And make some money in the process?
We’re in it this to make money and keep financing the R&D required to push the industry forward. We have to think commercially. It’s not just about developing technologies, we want to develop technologies that can dramatically grow and expand the industry in the near term, reducing cost and improving operations.