---
license: mit
tags:
  - humanoid-robotics
  - motion-imitation
  - reinforcement-learning
  - phc
  - claims
---

# CLAIMS PHC Assets

This repository hosts external assets for the CLAIMS project:

- intermediate CLAIMS PHC checkpoints
- one PHC baseline checkpoint
- PHC evaluation datasets used in our experiments

Related code repository:

- `https://github.com/WesleyXu224/CLAIMS_Code`

Related project page:

- `https://wesleyxu224.github.io/CLAIMS/`

## Directory Structure

```text
checkpoints/
  claims_single_prim/
    Humanoid_00014200.pth
    Humanoid_00031800.pth
    Humanoid_00050300.pth
    Humanoid_00061200.pth
    Humanoid_00074800.pth
    Humanoid_00086200.pth
    Humanoid_00094000.pth
  baseline_single_prim_amass/
    Humanoid.pth
sample_data/
  aist++_clean.pkl
  emdb_clean.pkl
  kungfu_g_clean.pkl
  video_convert_clean.pkl
```

## Asset Summary

`checkpoints/claims_single_prim/`

- CLAIMS PHC checkpoints saved at multiple training stages

`checkpoints/baseline_single_prim_amass/Humanoid.pth`

- baseline PHC checkpoint used for comparison

`sample_data/`

- PHC evaluation datasets in `pkl` format

## Usage

Example evaluation command:

```bash
python phc/run_hydra.py \
  learning=im_big \
  exp_name=claims_single_prim \
  env=env_im \
  robot=smpl_humanoid \
  env.motion_file=/path/to/video_convert_clean.pkl \
  env.num_envs=173 \
  test=True \
  epoch=94000 \
  headless=True \
  im_eval=True
```

Replace `env.motion_file` and checkpoint paths with your local paths after download.

## Notes

- These files are released separately from the code repository because of size.
- SMPL / SMPL-X assets are not included here.
- Please consult the main CLAIMS repository for the full pipeline and environment setup.