Reliable Robotics is partnering with Daedalean to develop next generation flight automation system technologies, the two companies revealed in an exclusive first interview to Avionics International.
Reliable Robotics, coming off its groundbreaking flight demonstration of a remotely piloted modified Cessna 208 Caravan last month, is partnering with Switzerland-based startup Daedalean to disrupt the concept of automated navigation and situational awareness technology inside next generation fixed and rotary-wing aircraft.
Daedalean CEO Luuk van Dijk and Reliable Robotics CEO Robert Rose, co-founder and CEO of Reliable Robotics, both former SpaceX engineers, provided emailed statements to Avionics International about the March 22 reveal of their new partnership.
"The end product both companies foresee is a system that can operate in airspaces as a model citizen, that enables denser economic use of the airspace, at safety levels that are an order of magnitude above today's standards," van Dijk said.
During their February flight test, Reliable Robotics remotely piloted a Cessna 208 Caravan from a control center in their Mountain View, California headquarters over 50 miles away from where the flight occurred. The company remotely piloted a modified Cessna 172 Skyhawk over a populated region with no one on board in 2019, and subsequently demonstrated fully automated landing of the larger Cessna 208 in 2020 on the third day of flight testing.
A newly released video published by Daedalean, embedded below, gives an overview of how their neural network visual guidance system works for aircraft navigation.
Daedalean has developed machine learning applications based on comprehensive situational awareness that meet aviation safety level standards defined by the Federal Aviation Administration (FAA) and European Union Aviation Safety Agency (EASA). Their work with EASA examining high performance machine learning algorithms for safety-critical applications resulted in the joint publishing of Concepts of Design Assurance for Neural Networks.
Now, Daedalean's advancements in machine learning algorithms and Reliable Robotics' autonomous aircraft system development will be paired to build advanced navigation and situational awareness systems for commercial aircraft operations.
“We take high-resolution video input from which we extract in real-time the most possible detail before sending it through a Convolutional Neural Network (CNN) tuned for the task at hand. We have demonstrated subsystems that can navigate, perform guidance to runways and safe emergency landing spots, and traffic detection,” van Dijk said.
An overview of how Daedalean uses neural convolutional neural network (CNN) technology for aircraft navigation and visual awareness.
“Using different helicopters and fixed-wing airplanes we demonstrated that we could meet high performance metrics in several different environments: our GPS-free navigation can stay in a 32m vertical corridor at 1500m [above ground level] AGL flight.”
Through the new partnership, Daedalean’s systems can now feed this information about the aircraft position relative to the terrain with its obstacles and safe landing sites and relative to other traffic, to the Reliable Robotics flight control stack, which then can use that information to deal with a variety of challenging tasks pilots can encounter, such as jammed or disabled GPS, non-cooperative traffic, or emergency landing scenarios, according to van Dijk.
Rose, the CEO and co-founder of Reliable Robotics, said that the company's remote pilot control center was designed by engineers who had worked on SpaceX's Falcon 9 and Dragon 2 programs.
Reliable Robotics CEO Robert Rose.
There, remote pilots communicate with ground and air traffic controllers as well as other aircraft using push-to-talk functionality.
"Each workstation includes a large monitor and an iPad that provides mission critical information for the remote pilot to manage the flight plan, and maintain situational awareness in order to instruct aircraft during all phases of flight," Rose said. "Command and control, voice and data links implement end-to-end encryption with authenticity and integrity checking to verify all data relayed between the remote pilot and the aircraft as original and unaltered."
Reliable Robotics remotely piloted this modified Cessna 208 aircraft with no human pilot onboard last month. (Reliable Robotics)
Along with their shared experience at SpaceX, the two startups have also placed a heavy focus on establishing government-industry consensus on the basis of engineering data and proof-of-concepts required to provide a path to safety critical certification of situational awareness and navigation systems that feature pilotless cockpits and use machine learning algorithms to update the aircraft’s flight path and avoid aerial obstacles or hazards.
Over the last year, Daedalean has also gone a step further to continue its advancement of the integration of machine learning into next generation flight control technology by providing experimental development kits to a number of undisclosed manufacturers.
"There is not a lot of choice in airworthy high-resolution digital cameras," van Dijk said. "Those available are prohibitively expensive or export controlled. Then, given that nobody really had to do this intense image processing and neural network computation on airworthy computers, there is also a need for certified processing platforms that exceed the state of the art."
“While we are working on closing that gap, we have started shipping an Evaluation Kit, based on commercial off-the-shelf components, available for selected partners to evaluate and experiment with our systems and to demonstrate that the functionality and software are ready and perform," he said.
Inside the kit are two cameras, one fisheye looking down and one wide-angle looking forward, that are paired with a processing box and a tablet as a ready-to-use user-interface.
Daedalean's evaluation kit allows aircraft makers to experiment with their technology.
Daedalean so far has focused on visual situational awareness, showing how modern machine learning techniques can be used to enhance and replace piloting tasks that are currently the exclusive domain of human pilots.
“While such neural-network architectures are becoming increasingly commonplace outside aviation, what Daedalean has focused on is doing this with the proof of safety and fitness-for-purpose required for airworthiness certification in mind,” van Dijk said.
"While developing the functionality further and extending its operational and environmental envelope is ongoing, our main focus right now is to work with the regulators to establish means-of-compliance to certify our Machine Learning based systems," he said. "We expect major updates in the course of this year. After that, and the hardware becoming available, nothing stands in the way of a broader deployment, beyond experimental applications, in the course of 2022.”