Abstract
This research aims to create open-source autonomous drone software to transform rivers into 21st century drone highways. We will explore vision-based navigation on ultra-light drones, with offloading of compute-intensive processing over wireless links to ground-based infrastructure. A major challenge is the presence of numerous bridges, whose large metallic structures distort wireless signals and lead to unreliable GPS-based navigation. Overcoming this will be the focus of our research.
Description
Rivers have served as transportation corridors for goods and people since the dawn of civilization. Today, they offer promise as natural safe corridors for autonomous drone flights. By using rivers as corridors, the flight paths of large, load-bearing drones over heavily populated urban areas can be kept to a minimum, thus improving public safety. Pittsburgh has the good fortune to have airspace over rivers. These interconnect communities and could become an ideal proving ground for drone transportation. We aim for fully autonomous flight: i.e., pre-programmed flight without a remote human pilot, including mission-specific actions in response to runtime observations.
A major challenge to autonomous drone flight over rivers are the numerous bridges. These large metallic structures distort wireless signals, and lead to unreliable GPS-based navigation. There is no human pilot to use his or her acute vision and intelligence to guide the drone. How does an autonomous drone avoid these obstacles reliably and safely?
For task autonomy, greater intelligence correlates with more powerful on-board computing and richer sensing. However, drones can only carry a limited payload. This is unlike an autonomous road vehicle that can easily carry LiDAR, multiple video cameras, and all the necessary compute capability. The weight of these sensing and computing entities reduces the amount of useful payload that can be carried. By reducing this weight to a minimum, we can optimize the working payload and hence the economic viability of drone-based transportation.
The key to overcoming on-board sensing and processing limitations is to offload intensive processing to ground-based computing infrastructure over a wireless link. Carnegie Mellon University pioneered the offloading concept, and has more recently pioneered its descendant technology of wireless edge computing.
In preliminary research, we have been exploring the use of an ultra-light drone for task autonomy. Our goal is to develop the software and algorithms needed for task autonomy on these lightweight platforms and perfect them, before advancing to heavier drones with substantial payload lift. From the viewpoint of public safety, small and light-weight drones are much more attractive than large drones. In case of catastrophic failure, the kinetic energy of a small drone is much less than that of larger and heavier drones. Inital experimentation with ultra-light drones to develop the software for navigation and obstacle avoidance is thus a prudent approach.
Our commercial off-the-shelf (COTS) drone weighs 320 g, inclusive of gimbal-mounted video camera. It uses a 3D-printed harness weighing 14 g to hold a COTS compute and 4G LTE network payload that weighs 26 g. Of the 320 g of total drone weight, 128 g is battery weight, giving a flight time of over 20 minutes. We believe that this is sufficient (pending experimental validation) to fly from Hazelwwood Green to the Point in Pittsburgh, including maneuvers to avoid bridges.
In summary, our goal is to create the open-source software for drone autonomy that transforms rivers into 21st century drone highways.
Timeline
December 1, 2022: Start of this research.
May 1, 2023: Completion of initial algorithms and coding.
June 15, 2023: Completion of experiments to validate algorithms.
June 30, 2023: Completion of result analysis and writeup.
Strategic Description / RD&T
Deployment Plan
We will conduct experiments at Mill-19 with live drone flights to validate the ability to navigate the Hot Metal Bridge.
Expected Outcomes/Impacts
The primary expected accomplishment is (a) release of source code for vision-based avoidance of bridge obstacles by drones, and (b) experimental confirmation of the success of this code in enabling river navigation. Secondary metrics include insights into the limitations of these algorithms, and insights into pathways for improvement through additional research.
Expected Outputs
TRID
Individuals Involved
| Email |
Name |
Affiliation |
Role |
Position |
| mbala@andrew.cmu.edu |
Bala, Mihir |
CSD |
Other |
Student - PhD |
| satya@cs.cmu.edu |
Satyanarayanan, Mahadev |
CSD |
PI |
Faculty - Tenured |
Budget
Amount of UTC Funds Awarded
$100000.00
Total Project Budget (from all funding sources)
$100000.00
Documents
| Type |
Name |
Uploaded |
| Data Management Plan |
2022_DMP_Bridge_Avoidance_in_River-based_Drone_Autonomy.pdf |
Dec. 20, 2022, 6:15 a.m. |
| Publication |
N/A |
March 27, 2023, 9:47 a.m. |
| Presentation |
N/A |
March 27, 2023, 9:47 a.m. |
| Progress Report |
411_Progress_Report_2023-03-30 |
March 27, 2023, 11:33 a.m. |
| Publication |
Does Wearable Cognitive Assistance Require Edge Computing? |
April 10, 2023, 9:34 p.m. |
| Publication |
Mega-nerf: Scalable construction of large-scale nerfs for virtual fly-throughs |
April 10, 2023, 9:35 p.m. |
| Publication |
The role of edge offload for hardware-accelerated mobile devices |
April 10, 2023, 9:35 p.m. |
| Final Report |
Final_Report_411.pdf |
July 26, 2023, 11:16 a.m. |
Match Sources
No match sources!
Partners
| Name |
Type |
| Department of Mobility Initiatives, Pittsburgh |
Deployment Partner Deployment Partner |
