#126 Certifying Autonomous Flight
Subscribe to get the latest
on 2023-02-23 08:00:00 +0000
with Darren W Pulsipher, Luuk Van Dijk,
In this episode, Luuk Van Dijk, CEO of Daedalean, talks with Darren about how his company has developed a technique to help governments certify AI-empowered autonomous flight in a highly regulated industry.
Keywords
#artificialintelligence #ai #machinelearning #policy
The aerospace industries are highly regulated, with some unique challenges in introducing AI-enabled autonomous flight. Because of the nature of flying, where things that go up can come crashing down and cause harm to people on board and people on the ground, regulations and certifications in this industry are very prescriptive and rigid. Luuk discusses the challenges and the techniques used to overcome those challenges in this episode of “Embracing Digital Transformation.”
Simplification
Electrification made everything much more accessible and straightforward. Everyone was expecting that urban air taxis would already be running now. However, regulations and certification that protect the public have slowed down the industry. People in the aerospace industry understand the challenge of working in this highly regulated space. Mavericks are trying to make this industry move faster, but government regulators are concerned about safety which slows things down. Regulations built around human pilots must change to enable machine flying. But these regulations have worked in the past and served the community well. So, there is resistance to change.
Regulatory Comparison
Companies like Daedalean are working with the government to certify AI and autonomous flight within the existing regulations and in some cases updating the current laws to allow this new mode of transportation while keeping the public safe.
Some surmise that the autonomous automobiles are like autonomous flight, but the design spaces are very different. Autonomous vehicles are less regulated than autonomous aircraft. Additionally, the physics of automobiles are fundamentally different. If a car gets confused, it can merely pull over to the side of the road. When an aircraft gets confused, it must land, which is the hardest part of flying. Additionally, autonomous vehicles have many more objects to avoid and have a complex operating environment, including people, objects, animals, traffic signs, and signals. In autonomous flight, the working environment is in 3D but does not have all the potential obstacles to avoid.
Complex Training
These complex environments require millions of hours of training for the AI algorithms for full autonomy. Additionally, regulators and the public need to gain confidence in AI’s ability autonomous fly aircraft. One of the approaches companies leverage is to augment safety by giving pilots an autonomous co-pilot. This is especially important in small aircraft and helicopters, where the safety numbers are comparable to motorcycle riding. Larger aircraft are already highly safe with incredible safety records.
Certifying AI algorithms.
In addition to certifying autonomous flight, many regulators are concerned about the ability to certify AI algorithms. This is an inherent problem in all AI algorithms that rely on extensive training sets to fine-tune the AI models. This comes from the fact that many regulators want to inspect the specific algorithms to understand how decisions are made for the algorithm. Many AI neuro networks used for pattern matching and visual AI processing are not defined by a simple mathematical model but instead fine-tuned by training data.
Building Trust with AI
AI tools like ChatGPT are showing how the masses can use AI. However, many people are worried about the accuracy of these AI algorithms in producing results. Luuk reminds us that AI is only as good as the data they are trained with. Garbage in, Garbage out. One of the problems is that people are beginning to trust tools like ChatGPT to give accurate results completely, which is misplaced trust. This could damage to trust needed for AI’s driving or flying autonomous vehicles.
More information about AI autonomous Flight can be found at https://www.daedalean.ai/.