Automated drone navigation for urban data collection

Yu, Tianhao

Permalink

https://hdl.handle.net/2142/127153 Copy

Description

Title

Automated drone navigation for urban data collection

Author(s)

Yu, Tianhao

Issue Date

2024-09-20

Director of Research (if dissertation) or Advisor (if thesis)

Caesar, Matthew

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Drone Visual Navigation
• Drone Path Planning

Abstract

In the development of smart cities, drone swarms are emerging as a cost-effective, automated solution for gathering essential urban data. In this task, the drone swarms are required to visit a set of sites with valuable urban data. However, many challenges exist in the site-visiting tasks of drone swarms: wide cities with complex streets and obstacles require drones to perceive the environment, avoid collisions, and plan paths with their limited onboard computation and battery. In situations where battery limitations often prevent visiting all intended sites, decisions must be made on which sites to visit as waypoints during the task so that the amount of data collected can be maximized. Moreover, the value of data sites may fluctuate due to dynamic urban conditions. For instance, disasters and emergencies may enhance the importance of immediate data scanning, significantly increasing the value of certain sites. Given all these challenges, intelligent planning is essential to prioritize which sites to visit dynamically and maximize the amount of data collected under these constraints. In this thesis, we address the following question: is it possible to develop a novel approach to enhance drone swarm navigation for data site visits and data collection in urban areas, despite constraints including limited onboard resources and dynamically changing environments? We introduce our solution, D-Planner, an architecture that performs efficient drone swarm waypoint planning, carries out collision-free navigation using a combination of computer vision and pathfinding modules, and integrates incremental real-time path recomputation that leverages computer vision and probabilistic navigation to figure out region data values for dynamic data collection decisions. D-Planner is structured around three key insights. First, the need to visit as many valuable data sites as possible can often be modeled as a waypoint planning problem in the context of drones. This is essential in contexts where global optimal computation is unfeasible due to drone limitations. Second, computer vision and pathfinding can be combined to navigate drones safely and effectively through urban regions. Lastly, we can leverage an incremental path recomputation module to dynamically incorporate new data sites into existing plans and accelerate path recomputation during drone operations. To evaluate the approach, we constructed a simulation environment of Manhattan, New York, with meter-level precision using Google Maps. Overall, D-Planner visits 25.60% more waypoints within 6.5× faster time compared with baseline algorithms. With real-time incremental path recomputation, D-Planner further discovers 21.8% more data sites during flights. Moreover, it improved the reaction time for obstacle avoidance by 56.14% compared with baseline algorithms. These observations suggest D-Planner’s great potential in planning data site visits and calculating collision-free trajectories in real time for drone swarm urban data collection.

Graduation Semester

2024-12

Type of Resource

Thesis

Handle URL

https://hdl.handle.net/2142/127153

Copyright and License Information

Copyright 2024 Tianhao Yu

Owning Collections