Learning a Controller Fusion Network by Online Trajectory Filtering for Vision-based UAV Racing

Abstract

Autonomous UAV racing has recently emerged as an interesting research problem. A common approach is to learn an end-to-end policy that directly predicts controls from raw images by imitating an expert. However, such a policy is limited by the expert it imitates and scaling to other environments is difficult. In this paper, we propose learning an optimized controller using a DNN that fuses multiple controllers. The network learns a robust controller with online trajectory filtering, which suppresses noisy trajectories and imperfections of individual controllers. The result is a network able to learn a good fusion of filtered trajectories from different controllers leading to significant improvements in overall performance. Our network beats all baselines in extensive experiments and approaches the performance of a professional human pilot.

Resources

arXiv: 1904.08801

Video

Citation

@inproceedings{mueller2019cfn,
  title     = {Learning a Controller Fusion Network by Online Trajectory Filtering for Vision-based {UAV} Racing},
  author    = {M{\"{u}}ller, Matthias and Li, Guohao and Casser, Vincent and Smith, Neil and Michels, Dominik L. and Ghanem, Bernard},
  booktitle = {CVPR Workshops (UAVision)},
  note      = {Oral},
  year      = {2019}
}