Implementation of Partial Observable Markov Decision Process (POMDP) algorithm using Bitcraze Crazyflie Drones

Abstract

This paper develops a complex navigation solution for sequential drone swarm configuration using Partially Observable Markov Decision Process (POMDP) with the Bitcraze Crazyflie platform with a single localisation anchor. The objective being to generate a stable control system for a swarm of drones to navigate a controlled environment towards a waypoint. The POMDP solver takes observations of the drones’ real-world positions and determines specific actions based upon a network of functions designed to optimise a path towards the waypoint. Once the solver defines the next action the swarm navigates towards the selected direction sequentially. Through extensive developmental and formal testing, the developed system performs the objective with an average trajectory deviation of less than 0.1 meters with a duration of approximately 18 seconds. Deficiencies have been identified in the software control structure. This research highlights the importance of drone control and localisation redundancies for complex navigation solutions for micro-UAV swarm configurations.