Video Prediction Policy 2: Predict Better, Act Better
Official VPP2 weights for zero-shot video prediction, RoboDojo and LIBERO.
Paper · Code · Project page · ModelScope
Model weights
| Checkpoint | Path | Use |
|---|---|---|
| VPP2 Stage-1 Video (49 frames) | checkpoints_video/vpp2-video-stage1-49f.pth | Zero-shot image-to-video prediction |
| VPP2 Stage-2 Video (17 frames) | checkpoints_video/vpp2-video-stage2-17f.pth | Zero-shot image-to-video prediction |
| RoboDojo history-conditioned Video-10k | checkpoints/initialization/robodojo_his10k.pt | Joint + Action2B training initializer |
| RoboDojo joint Video-100k | checkpoints/joint2b_s100000/video.pt | Paired RoboDojo evaluation |
| RoboDojo Action2B-100k | checkpoints/joint2b_s100000/action.pt | Paired RoboDojo evaluation |
| LIBERO Video-10k | checkpoints/libero/video_step010000.pt | Action training and evaluation |
| LIBERO Action-30k | checkpoints/libero/action_step030000.pt | LIBERO, LIBERO-OOD and LIBERO-PRO evaluation |
Shared VAE, CLIP, UMT5 and tokenizer assets are under
checkpoints/Wan2.1-I2V-14B-480P/. Each policy pair includes its manifest and
normalization statistics. Keep each pair together.
Download
Install the Hugging Face CLI:
python -m pip install -U huggingface_hub
Each command downloads the root config.json, which lists the
released checkpoints, video frame counts and inference defaults, together with
the selected weights and shared Wan encoders.
Stage-1 video
Use --num-frames 49.
hf download Haodong082399/VPP2 --local-dir weights \
--include 'config.json' \
--include 'checkpoints_video/vpp2-video-stage1-49f.pth' \
--include 'checkpoints/Wan2.1-I2V-14B-480P/**'
Stage-2 video
Use --num-frames 17.
hf download Haodong082399/VPP2 --local-dir weights \
--include 'config.json' \
--include 'checkpoints_video/vpp2-video-stage2-17f.pth' \
--include 'checkpoints/Wan2.1-I2V-14B-480P/**'
RoboDojo training
Use Video-10k to initialize joint Video + Action2B training.
hf download Haodong082399/VPP2 --local-dir weights \
--include 'config.json' \
--include 'checkpoints/initialization/robodojo_his10k.pt' \
--include 'checkpoints/Wan2.1-I2V-14B-480P/**'
RoboDojo evaluation
Download the paired Video-100k and Action2B-100k weights with their normalization statistics.
hf download Haodong082399/VPP2 --local-dir weights \
--include 'config.json' \
--include 'checkpoints/joint2b_s100000/**' \
--include 'checkpoints/Wan2.1-I2V-14B-480P/**'
LIBERO
Use the same Video-10k and Action-30k pair for LIBERO, LIBERO-OOD and LIBERO-PRO.
hf download Haodong082399/VPP2 --local-dir weights \
--include 'config.json' \
--include 'checkpoints/libero/**' \
--include 'checkpoints/Wan2.1-I2V-14B-480P/**'
Both standalone video checkpoints use 30 denoising steps, CFG=4 and sigma shift 3. Follow the video prediction guide for image preparation and inference, or the RoboDojo and LIBERO guides for policy training and evaluation.
All weights
To download the complete repository, including config.json:
hf download Haodong082399/VPP2 --local-dir weights
Evaluation settings
| Benchmark | Checkpoint pair | Denoising | Action horizon / execution |
|---|---|---|---|
| RoboDojo | Video-100k + Action2B-100k | 10 Euler steps, shift 1 | 32 / 24 |
| LIBERO / OOD / PRO | Video-10k + Action-30k | 10 Euler steps, shift 5 | 32 / 10 |
RoboDojo uses its EE16 z-score statistics; LIBERO uses its own 7D action min/max statistics. Use each benchmark's configuration and simulator setup.
Citation
See CITATION.bib for the paper citation.
Licenses
Shared Wan assets retain their upstream license in
checkpoints/Wan2.1-I2V-14B-480P/LICENSE.txt.
The official implementation has its own code license.