CrazyKhoreia: A Robotic Perception System for MAV Path Planning From Digital Images

Abstract Micro Air Vehicles (MAVs) can be used for a wide range of applications, such as drone-based light shows for cultural or marketing purposes, or as a replacement for fireworks. However, more widespread usage of this technology is precluded by the usability gap between choreography design and deployment on target drones. Seamless deployment of MAV choreographies requires the building of computational interfaces that can obtain drone trajectories from conventional media such as digital images. In this paper, we propose CrazyKhoreia, a low-cost, computationally efficient approach to MAV choreography design. CrazyKhoreia is a robotic perception system that obtains a safe, traversable, and accurate waypoint matrix from a digital image. We validate our trajectory generation system through two distinct modes of operation: light painting, where an MAV flies through all waypoints, and multi-drone formation, in which multiple MAVs are arranged to emulate a given image. The utility of each mode is evaluated differently, using the full resolution CrazyKhoreia output as a reference for comparison. For the light painting mode, we logged the MAV’s pose at a frequency of 10 Hz and compared it with the reference for different levels of detail in the reproduced path. The Root Mean Squared Error (RMSE) ranges from 0.1287 to 0.3190 m. For multi-drone formation, where the swarm remains stationary, we computed the RMSE by comparing observed positions with the reference across multiple tests, resulting in a mean RMSE value of 0.1360 m.