Publish PredActor evaluation artifacts

README.md

@@ -0,0 +1,82 @@
+---
+license: other
+library_name: pytorch
+tags:
+- robotics
+- humanoid
+- unitree-g1
+- diffusion-policy
+- mujoco
+- motionclip
+---
+
+# PredActor Artifacts
+
+Official evaluation artifacts for
+[PredActor](https://github.com/MasterYip/PredActor), a predictive action
+diffusion controller for steerable Unitree G1 locomotion.
+
+This repository contains only the two learned artifacts required by the public
+MuJoCo evaluation path. Source code and robot assets live in the PredActor code
+repository; datasets, experiment outputs, deployment bundles, and additional
+checkpoints are not included here.
+
+## Repository layout
+
+```text
+PredActor_Artifacts/
+├── README.md
+├── SHA256SUMS
+├── checkpoints/
+│   ├── predactor/
+│   │   └── pdp051/
+│   │       └── latest.ckpt
+│   └── motionclip/
+│       └── g1-model-xyz-clip/
+│           └── checkpoint_0100.pth.tar
+└── dataset/
+    └── README.md
+```
+
+`dataset/` is reserved for future public datasets and currently contains no
+dataset payloads.
+
+| Artifact | Size | SHA-256 |
+|---|---:|---|
+| PredActor PDP051 policy | 49,212,894 bytes | `2d963b32786f2989c6472726df9fcfe6b385590127e12e1f549a4b7d77488b2e` |
+| G1 MotionCLIP encoder | 542,749,069 bytes | `66a127df4958b346089b2020f2705c7456d9db0ee8b4bd9518608b708b35fc3c` |
+
+`SHA256SUMS` provides the same identities in a format accepted by
+`sha256sum --check`.
+
+## Download
+
+From a clone of the PredActor code repository:
+
+```bash
+python -m pip install huggingface_hub pyyaml
+python scripts/hf_artifacts.py download
+```
+
+The default destination is a sibling `PredActor_Artifacts/` directory, keeping
+learned weights outside the Git checkout. To choose another location:
+
+```bash
+python scripts/hf_artifacts.py --local-root /absolute/artifact/path download
+```
+
+The downloader verifies byte size and SHA-256 before reporting success.
+
+## Evaluation
+
+Pass the downloaded paths to `PredActor/predactor_eval.py` together with the
+hashes above. The evaluator intentionally refuses learned weights stored inside
+the source checkout.
+
+## Scope and terms
+
+These checkpoints are provided for research evaluation. The repository uses
+`license: other` because the artifacts and their transitive components do not
+share one standalone license grant. Users are responsible for complying with
+the terms of the PredActor source, MotionCLIP/OpenAI CLIP dependencies, Unitree
+assets, and any applicable third-party software.

SHA256SUMS

@@ -0,0 +1,2 @@
+2d963b32786f2989c6472726df9fcfe6b385590127e12e1f549a4b7d77488b2e  checkpoints/predactor/pdp051/latest.ckpt
+66a127df4958b346089b2020f2705c7456d9db0ee8b4bd9518608b708b35fc3c  checkpoints/motionclip/g1-model-xyz-clip/checkpoint_0100.pth.tar

checkpoints/motionclip/g1-model-xyz-clip/checkpoint_0100.pth.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66a127df4958b346089b2020f2705c7456d9db0ee8b4bd9518608b708b35fc3c
+size 542749069

checkpoints/predactor/pdp051/latest.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d963b32786f2989c6472726df9fcfe6b385590127e12e1f549a4b7d77488b2e
+size 49212894

dataset/README.md

@@ -0,0 +1,4 @@
+# Datasets
+
+This directory is reserved for future PredActor dataset releases. No dataset
+payloads are published in the current artifact release.