FlowPilot Checkpoints

Model checkpoints for FlowPilot, a Flow Matching based world-action model for real-time onboard UAV navigation.

The model uses a dual-stream MoT to jointly model video latents and action tokens. Actions are parameterized by a 7th-order Bernstein polynomial with 8 control points (3 fixed by the initial state and 5 predicted by the network), producing 48 waypoints at 30 Hz.

• Code: flyingbitac/FlowPilot

Files

File	Size	Description
ckpt/phase2/ckpt_final.pt	651 MB	Phase-2 checkpoint. The inference state dict is stored under the model key; step, optimizer, and lr_scheduler are also included.
ckpt/action_stats.pt	3.1 KB	Action and state normalization statistics.
ckpt/coeff_stats.pt	2.7 KB	Bernstein coefficient normalization statistics.

The Wan2.2 VAE weights are not stored in this repository; download the official file from Wan-AI/Wan2.2-TI2V-5B as shown below.

Usage

Clone the code repository first, then download the checkpoints into its root:

git clone https://github.com/flyingbitac/FlowPilot.git
cd FlowPilot

hf download YuDinging/FlowPilot --include "ckpt/*" --local-dir .
hf download Wan-AI/Wan2.2-TI2V-5B Wan2.2_VAE.pth --local-dir ckpt

If hf is not available, install it with pip install -U huggingface_hub.

The --include "ckpt/*" filter matters: without it, hf download would also write this model card as README.md and overwrite the README of the code repository.

The files land under ckpt/, the default location expected by configs/runtime/rviz.py:

ckpt/
├── phase2/ckpt_final.pt
├── Wan2.2_VAE.pth
├── action_stats.pt
└── coeff_stats.pt

hf download also creates a .cache/huggingface/ directory in the current directory. It only holds download metadata and can be removed afterwards. If you prefer to keep the weights outside the repository, download to another directory with --local-dir and symlink its ckpt/ into the repository instead.

Then start the simulation from the code repository:

python docker/docker.py build
python docker/docker.py start

Intended Use

These weights are released for research on learning-based UAV navigation, mainly for simulation and onboard inference experiments. They are not certified for safety-critical flight.

License

The checkpoints are released under the Apache License 2.0.

ckpt/Wan2.2_VAE.pth is downloaded from the official Wan2.2 model repository (Alibaba Wan Team) and is also licensed under Apache-2.0.