Autonomous landing of Micro Air Vehicles through bio-inspired monocular vision
Hann Woei Ho · Research Repository (Delft University of Technology) · 2017
Autonomous flying vehicles -especially the small and light-weight Micro Air Vehicles (MAVs) -are currently not capable of finding a safe landing site by themselves and performing smooth landings.The main reason is that most of the state-of-the-art solutions are computationally expensive and thus restrict their application on MAVs which have severely limited processing power and sensors available.Hence, this kind of vehicles requires an energy-efficient landing solution using a light-weight sensor.A promising option for MAV landing is to draw inspiration from tiny flying insects.The motivation comes from the fact that these insects have very limited neural and sensory resources, yet they can perform complex tasks, such as navigating in a dense, obstacle-rich environment and carrying out smooth landings.This means that they possess efficient and robust approaches to perception and control problems, and these approaches could be ideal for designing the control strategies for MAV landing.Research shows that, to realize many of these complex tasks, flying insects use control strategies which heavily rely on optical flow.However, there are many challenges when using solely optical flow for MAV landing.Optical flow is 'scaleless' and does not provide the distance to an object or the observer's velocity, but only the ratio of them.The scaleless property of optical flow can lead to various stability problems of MAV landings.For instance, for landing on a flat surface, optical flow control causes instabilities at some points in the phase of the landing when the control gains are not adapted to the height.Thus, height information is important in optical flow landings.Besides, for a fully autonomous MAV, being able to land on a flat surface is not sufficient.The MAV must also be able to find a suitable landing site by itself.In fact, optical flow information obtained from a single camera is sparse and needs to be further interpreted to understand what the flow pattern actually represents.Therefore, MAVs need efficient algorithms which can analyze the optical flow to search for a ground surface that is relatively flat and also -perhaps even more important -free of obstacles.vii