---
license: mit
library_name: pytorch
---

# Push-T diffusion policy checkpoints

Two Push-T diffusion policy checkpoints with a minimal inference script.

## Files

- `original.ckpt` — original checkpoint
- `edited.pt` — edited checkpoint
- `stats_sfp.npz` — normalization stats (obs / action min/max)
- `inference.py` — inference script

## Quick start

```bash
pip install torch diffusers pygame pymunk shapely scikit-image imageio gym numpy
git clone https://github.com/columbia-ai-robotics/streaming_flow_policy.git

python inference.py --ckpt original.ckpt --stats stats_sfp.npz \
    --sfp-repo streaming_flow_policy/ --n-seeds 50

python inference.py --ckpt edited.pt    --stats stats_sfp.npz \
    --sfp-repo streaming_flow_policy/ --n-seeds 50 --save-mp4 rollout.mp4
```

Prints success rate and mode distribution; optional `--save-mp4` renders the first successful rollout.