📡 Embodied AI Radar具身智能雷达 Star2

Most of this project is curated by Claude. Drop a star to cheer Claude on!本项目大量由 Claude 整理维护,点个 Star 给克劳德大小姐鼓励一下!

★ Feed a star投喂 Star

Topics主题 › Navigation & Embodied Agents导航与具身智能体

Paper论文 2026 new新 2026-11-04

Aero-VPA: Few-Step Flow-Matching Control over Non-Reconstructive JEPA Latents for Agile Flight on Edge Hardware

A compact JEPA-plus-flow-matching quadrotor policy with an unusually candid report: cloning failed all 60 missions, DAgger plateaued, and a runtime shield raised simulated forest success from 21.1% to 52.2%.紧凑的 JEPA 加流匹配四旋翼策略,报告异常坦率:行为克隆 60 次任务全败,DAgger 进入瓶颈,运行时安全层将仿真森林任务成功率从 21.1% 提至 52.2%。

BibTeX
@misc{lee2026aero,
  title = {{Aero-VPA: Few-Step Flow-Matching Control over Non-Reconstructive JEPA Latents for Agile Flight on Edge Hardware}},
  author = {Yeonseok Lee},
  year = {2026},
  eprint = {2611.04392},
  archivePrefix = {arXiv},
  primaryClass = {cs.LO},
  url = {https://arxiv.org/abs/2611.04392},
}
Navigation & Embodied Agents导航与具身智能体Drone / aerial无人机simdronejepaflow-matching

Authors (from arXiv): 作者(来自 arXiv):Yeonseok Lee

Code license代码许可证 (detected by GitHub): (GitHub 检测):GPL-3.0

Similar entries相似条目

All Navigation & Embodied Agents entries →查看全部导航与具身智能体条目 →