DepDrone: Rigorously Investigating Drone Dependability

Important results from the project

DepDrone sought to overcome the inability to provide solid guarantees on the dependable operation of aerial drones. The project studied key issues such as i) testing the device reactions when pushing the drone to its physical limits, ii) understanding the impact of sensor failures, inaccuracies, and environment effects on perception functionality such as SLAM, and iii) understand how varying network conditions affect the dependability of drone operations.

Expected long term effects

Through a series of real-world tests in a special-purpose testbed able to accurately emulate given environment dynamics in a repeatable manner, we both tested the robustness of existing SLAM functionality against environment dynamics and hardware/software failures, and developed novel SLAM functionality is more robust against these issues. We established contacts with the Ardupilot developers to possibly integrate our SLAM functionality into the codebase, and expect that to be employed in challenging setups where dependable operation is key.

Approach and implementation

We customised the existing FlyZone testbed (bit.ly/33VE7Ya) to emulate the environment influence on SLAM-based functionality. We developed add-on SLAM functionality using new perception algorithms that take as input a custom-built navigation board that includes stereoscopic cameras and a purpose-built IMU. We created a hardware prototype and accumulated a total of 600h of test flights. We finally performed preliminary off-line tests of our novel SLAM functionality against algorithms found in the state of the art.