Approximate Imitation Learning for Event-based Quadrotor Flight in Cluttered Environments
Event-camera drone flight: learn the representation offline from a large dataset, then fine-tune the policy online on cheap state only, with no event rendering. Training drops from 52.44 h to 1.86 h.事件相机无人机飞行：先在大规模离线数据上学习表示，再只用轻量状态信息在线微调策略，无需在训练中渲染事件。训练时间从 52.44 小时降到 1.86 小时。
BibTeX
@misc{messikommer2026approximate,
title = {{Approximate Imitation Learning for Event-based Quadrotor Flight in Cluttered Environments}},
author = {Nico Messikommer and Jiaxu Xing and Leonard Bauersfeld and Marco Cannici and Elie Aljalbout and Davide Scaramuzza},
year = {2026},
eprint = {2603.07578},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2603.07578},
}Navigation & Embodied Agents导航与具身智能体Drone / aerial无人机event-cameraquadrotorimitation-learningagile-flight