This problem is compounded when the drone tries to land on a moving platform such as a delivery van or a warship pitching in high seas.
“It has to land within a designated area with a small margin of error,” Prof Kumar said. “Landing a drone on a moving platform is a very difficult problem scientifically and from an engineering perspective.”
To address this challenge, UC researchers have applied a concept called fuzzy logic, the kind of reasoning people employ subconsciously every day.
While scientists are concerned with precision and accuracy in all they do, most people get through their day by making inferences and generalities, or by using fuzzy logic. Instead of seeing the world in black and white, fuzzy logic allows for nuance or degrees of truth.
“In linguistic terms, we say large, medium and small rather than defining exact sets,” prof Kumar explained. “We want to translate this kind of fuzzy reasoning used in humans to control systems.”
Fuzzy logic helps the drone make good navigational decisions amid a sea of statistical noise. It's called ‘genetic-fuzzy’ because the system evolves over time and continuously discards the lesser solutions. It is currently being put to the test in experiments to land quadcopters on robots mounted with landing pads at UC's UAV Multi-Agent System Research (MASTER) Lab.
Kelly Cohen, UC aerospace engineering professor, is confident about the team's approach: “Compared to other state-of-the-art techniques of adaptive thinking and deep learning, our approach appears to possess several advantages. Genetic fuzzy is scalable, adaptable and very robust.”
This project could solve real-world problems, such as equipping a delivery vehicle with a companion drone that is able to make deliveries and land itself. The US military has also registered an interest after another UC doctoral graduate Nick Ernest, started an artificial intelligence company called Psibernetix, that demonstrated the power of a fuzzy-logic-based artificial intelligence, dubbed ALPHA, that bested a human fighter pilot in simulated dogfights.
