Electronic Thesis and Dissertation Repository

Thesis Format



Master of Science


Computer Science


Haque, Anwar


Unmanned aerial vehicles (UAVs) can provide automated approaches to remote data collection, inspection, and exploration. When doing so, these tasks are accompanied by commercial gains and safety benefits. While UAVs can automate these tasks in some environments today, their size, resource requirements, and communication dependencies prevent them from exploring many other types of environments. These barriers are especially prevalent in areas that deny wireless communication and are too small for larger-bodied UAVs. In this work we present a novel exploration planner capable of overcoming these barriers and operating in otherwise inaccessible environments. To achieve this, we present a new approach to frontier detection and mapping which enables exploration and scales to nano sized UAVs. We prove the viability of this solution through real-world experimentation at WING research labs. This exploration software accommodates the extreme resource constraints of the small UAVs required to fly in confined spaces. The presented strategy is truly autonomous with no dependency on communication with external systems and no prior knowledge of the exploration space. To the best of our knowledge, the presented prototype can explore the smallest spaces that have yet to be reached by connectionless and autonomous UAVS. This claim is supported by the demonstration of real-world testing as our prototype achieved full exploration of several challenging environments.

Summary for Lay Audience

Autonomous Unmanned Aerial Vehicles (UAVs) are a rapidly developing technology that is influencing many fields today. An exciting and commercially viable application of this technology is autonomous exploration and inspection of spaces that are inaccessible to humans. These spaces come in a variety of forms presenting varying constraints and affordances to the flight of drones. In some of the most constrained environments, we observe the denial of wireless communication between a drone and external systems. If access to the environment is not available before a mission, the UAV is required to make complicated navigational decisions onboard during flight. Such an environment can also feature narrow passages placing restrictions on the physical size of the drone. This in turn limits the computing power, battery size, and payload capacity available. The lighter weight also makes the drone less stable in windy conditions or when propellers cause gusts of air to reflect off nearby obstacles and back towards the drone. The ability to inspect small spaces despite these factors will provide new business opportunities while promoting health and safety. In eliminating the need for expensive human inspection, companies can automate inspection tasks. In doing so, they also eliminate human exposure to potentially hazardous working conditions.

In this work, we introduce strengths and weaknesses of various types of UAVs. We then examine components that enable autonomous navigation in previously unknown environments, from low level flight dynamics to path planning for exploration. By reviewing existing solutions for autonomous exploration, we distinguish the barriers preventing existing approaches from moving to smaller connectionless spaces. With our motivation to overcome these barriers, we present a novel exploration strategy that enables small-bodied UAVs to explore previously unknown, confined, and connectionless environments. To operate in this environment, we present a novel frontier-based exploration method capable of navigating in spaces less than 50cms wide. We demonstrate our design onboard a commercially available nano-sized UAV. We evaluate this prototype on multiple real world test beds imitating different types of environments. To the best of our knowledge, this prototype explores more confined environments than any other fully autonomous solution existing today.

Available for download on Saturday, August 31, 2024

Included in

Robotics Commons