Event
Zhejiang University's Gao Fei Team Releases CrossFly UAV Navigation Benchmark
1 reports · 1 sources
The Gao Fei team at Zhejiang University, in collaboration with Differential Intelligence Flying (Weifen Zhifei), has released the CrossFly benchmark, accompanied by a hierarchical VLA baseline, CrossFly-Agent; the full simulation dataset, evaluation protocol, and baseline code are all open-sourced. The benchmark contains 23 reconstructed real-world scenes and 15,263 complete trajectories, covering three task categories: indoor, outdoor, and indoor-outdoor cross-domain, with indoor-to-outdoor cross-domain trajectories accounting for the highest proportion at 51.2%; the final dataset contains 3 million frames, with a total flight duration of 160+ hours and a total distance of 120 km. The simulation foundation is based on NVIDIA Isaac Sim, with PhysX handling flight dynamics and 3DGS responsible for photorealistic rendering, supporting two embodiments: a quadrotor equipped with an SO100 robotic arm and a humanoid-hand-configuration flying robot. Experiments are divided into two major evaluations: seen and unseen scenes, using scene-level splits, where all trajectories from the same scene are assigned either entirely to the training set or entirely to the test set, avoiding information leakage.
Report timeline
- Zhejiang University's Gao Fei team releases CrossFly benchmark: evaluation of cross-indoor-outdoor interactive navigation for drones and a hierarchical VLA baseline
The Gao Fei team at Zhejiang University, together with Weifen Zhifei, launched the CrossFly benchmark, building 23 reconstructed real-world scenes and 15,263 complete trajectories, covering three task types: indoor, outdoor, and indoor-outdoor cross-domain, accompanied by a hierarchical VLA baseline, CrossFly-Agent; the full simulation dataset, evaluation protocol, and baseline code are all open source.
深蓝AI · 微信公众号 ·