Autonomous drone cinematographer: Using artistic principles to create smooth, safe, occlusion-free trajectories for aerial filming

Autonomous aerial cinematography has the potential to enable automatic\ncapture of aesthetically pleasing videos without requiring human intervention,\nempowering individuals with the capability of high-end film studios. Current\napproaches either only handle off-line trajectory generation, or offer\nstrategies that reason over short time horizons and simplistic representations\nfor obstacles, which result in jerky movement and low real-life applicability.\nIn this work we develop a method for aerial filming that is able to trade off\nshot smoothness, occlusion, and cinematography guidelines in a principled\nmanner, even under noisy actor predictions. We present a novel algorithm for\nreal-time covariant gradient descent that we use to efficiently find the\ndesired trajectories by optimizing a set of cost functions. Experimental\nresults show that our approach creates attractive shots, avoiding obstacles and\nocclusion 65 times over 1.25 hours of flight time, re-planning at 5 Hz with a\n10 s time horizon. We robustly film human actors, cars and bicycles performing\ndifferent motion among obstacles, using various shot types.\n