Imitation Drone Dodges Black Holes of Uncertainty express
When a drone learns from an experienced pilot, it memorizes not just maneuvers, but entire safe corridors. But in real flight, there are always factors absent in training: unexpected wind, camera shake, even the drone's own motion can destabilize the air. Ordinary imitation learning — mimicking a human mentor — fails here: under changed conditions, copying leads to a crash. This is how distribution shift manifests: what the drone learned in the simulator may turn out wrong in reality.
The new algorithm solves this problem by drawing invisible 'black holes' around all undesirable zones. Mathematically, they are described so that the closer the drone gets to danger, the stronger the system repels it — like reverse gravity. Thanks to this, the drone smoothly skirts forbidden areas, even if sensors 'lie' or communication breaks.
Tests showed the vehicle never breached the safety boundary, handling strong crosswinds and sudden failures. Such caution paves the way for urban delivery in any weather or rescue missions where a mistake costs lives.
Original abstract
as written by the authors arXiv:2607.13436
Imitation learning (IL) has achieved remarkable success in complex decision-making tasks. However, its performance is highly sensitive to distribution shifts, which can pose significant safety risks. We propose a distributionally robust and safe IL framework that explicitly addresses both policy-induced and uncertainty-induced distribution shifts. Our approach develops a unified framework leveraging Taylor Series Imitation Learning (TaSIL) to mitigate policy-induced shifts and distributionally robust adaptive control to handle uncertainty-induced shifts. This architecture enables the formulation of an IL problem that optimizes performance under distributional uncertainty while systematically accounting for safety constraints. We demonstrate the effectiveness of the proposed approach on an unmanned aerial vehicle (UAV) case study where the UAV performs a task in an uncertain environment while avoiding unsafe regions.
Taylor series in the algorithm work like a navigator: they replace a complex trajectory with a sequence of small straight segments.
In the movie 'Interstellar', the heroes maneuvered around the black hole Gargantua, relying on calculations and adapting to warped space — roughly how this drone operates, but in miniature.