As part of the EPSRC CASCADE Programme on aerial robotics, we studied semi-autonomous control for drones inspecting structures.
What kinds of structures?
As the name suggests, we set out to target bridges, and we were especially fortunate to try some early flight trials at the Clifton Suspension Bridge, by kind permission of the Clifton Suspension Bridge Trust. We were able to prove our ability to fly in the bridge environment (during a planned closure) and to detect the bridge using a simple, lightweight LIDAR sensor.
Unfortunately, rather a lot happened between 2018 and 2023, which meant we were unable to return to the bridge for a full trial of our final system. Later trials took place at the Snowdonia Aerospace Centre in Llanbedr, North Wales.
What is semi-autonomous control?
For us, it means keeping the pilot in control of the system, while automating appropriate tasks such as monitoring the vehicle health and checking distances. The drone remains within visual line of sight of the pilot, and the pilot remains vigilant for air risks (other aircraft) and ground risks (e.g. people or vehicles nearby).
This partnership demands a simple interface between the pilot and the control system. It must not become more complicated for the pilot to figure out what the control system is doing. For this reason, we adopted simple "traffic light" feedback from the control to the pilot, and a simple "yes/no" input from the pilot to the control. The concept is inspired in part by railway signalling systems.
Under the hood, we use a behaviour tree to create a modular controller, combining many different checks and actions in a flexible, re-usable way.
What research did we actually do?
- Developed and validated the concept
- Designed a set of behaviour tree templates to build different missions
- Showed how formal safety arguments could be constructed for this combination of human and automation
- Deployed the system in both immersive simulation and real flight
- Conducted a human factors trial of the system (results pending)
Where can I learn more?
- The original case study design is here. (We didn't do all of it...)
- A paper on the semi-autonomous drone control concept
- A paper on mission templates in behaviour trees
- A paper on safety analysis
- A paper on the immersive simulation deployment, in collaboration with CASCADE partners at the University of Manchester
- A good introduction to behaviour trees can be found in the book by Colledanchise and Ögren.
Can I try it myself?
Our Python library for control of drones is available on GitHub.
Acknowledgements
Research at the University of Bristol FlightLab by PhD student Hirad Goudarzi supported by technical specialist Duncan Hine. Supervised by Arthur Richards and Tom Richardson. Supported by the UK Engineering and Physical Sciences Research Council as part of the CASCADE programme grant EP/R009953/1.
