Downloads · 30 days
2
2% of all-time downloads
whatiname/lehome_submission_ply100
lehome_submission_ply100 is a machine learning model from whatiname. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Downloads · 30 days
2
2% of all-time downloads
All-time downloads
91
Public
Parameters
140M
558 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors558 MB · 100%
From the Hugging Face model README
<a name="english"></a>
This submission uses a custom ACT-XSA (Action Chunking Transformer with custom configuration) policy built on top of the LeRobot framework.
act_xsa)whatiname/lehome_submission_ply100LeHome Challenge Environment: Follow the official setup guide to install Isaac Sim, IsaacLab, and the challenge codebase.
LeRobot: Install the standard LeRobot package:
pip install lerobot
Custom Policy Package: The model uses a custom policy type (act_xsa). Install the provided policy package:
cd lerobot_policy_act_xsa
pip install -e .
After installing the custom policy package, add one line to scripts/eval_policy/lerobot_policy.py to register the act_xsa policy:
# In scripts/eval_policy/lerobot_policy.py, add this line with other imports:
import lerobot_policy_act_xsa # Register custom act_xsa policy
Why this is needed: The model's config.json uses "type": "act_xsa". This import registers act_xsa as a valid policy type in LeRobot's configuration system. Without it, you will get:
Couldn't find a choice class for 'act_xsa' in <class 'lerobot.configs.policies.PreTrainedConfig'>
After downloading from HuggingFace, the model directory contains:
pretrained_model/
├── config.json # Policy config (type: "act_xsa")
├── model.safetensors # Model weights
├── train_config.json # Training configuration
├── policy_preprocessor.json # Observation preprocessor
├── policy_postprocessor.json # Action postprocessor
├── policy_preprocessor_step_*.safetensors # Preprocessor weights
├── policy_postprocessor_step_*.safetensors # Postprocessor weights
└── lerobot_policy_act_xsa/ # Custom policy package (install this!)
├── pyproject.toml
└── src/
└── lerobot_policy_act_xsa/
├── __init__.py
├── configuration_act_xsa.py
├── modeling_act_xsa.py
└── processor_act_xsa.py
cd lehome-challenge
python -m scripts.eval \
--policy_type lerobot \
--policy_path /path/to/downloaded/pretrained_model \
--garment_type "top_long" \
--dataset_root Datasets/example/top_long_merged \
--num_episodes 5 \
--enable_cameras \
--device cpu
Available garment types: top_long, top_short, pant_long, pant_short
<a name="中文"></a>
本提交使用基于 LeRobot 框架的自定义 ACT-XSA(自定义配置的 Action Chunking Transformer)策略。
act_xsa)whatiname/lehome_submission_ply100LeHome Challenge 环境: 按照官方指南安装 Isaac Sim、IsaacLab 和挑战代码库。
LeRobot: 安装标准 LeRobot 包:
pip install lerobot
自定义策略包: 本模型使用自定义策略类型 (act_xsa),需要安装提供的策略包:
cd lerobot_policy_act_xsa
pip install -e .
安装自定义策略包后,在 scripts/eval_policy/lerobot_policy.py 中添加一行代码以注册 act_xsa 策略:
# 在 scripts/eval_policy/lerobot_policy.py 中,与其他 import 一起添加:
import lerobot_policy_act_xsa # Register custom act_xsa policy
为什么需要这行代码: 模型的 config.json 中使用的是 "type": "act_xsa"。这行 import 会将 act_xsa 注册为 LeRobot 配置系统中的有效策略类型。如果没有这行代码,会报错:
Couldn't find a choice class for 'act_xsa' in <class 'lerobot.configs.policies.PreTrainedConfig'>
从 HuggingFace 下载后,模型目录包含:
pretrained_model/
├── config.json # 策略配置(type: "act_xsa")
├── model.safetensors # 模型权重
├── train_config.json # 训练配置
├── policy_preprocessor.json # 观测预处理器
├── policy_postprocessor.json # 动作后处理器
├── policy_preprocessor_step_*.safetensors # 预处理器权重
├── policy_postprocessor_step_*.safetensors # 后处理器权重
└── lerobot_policy_act_xsa/ # 自定义策略包(必须安装!)
├── pyproject.toml
└── src/
└── lerobot_policy_act_xsa/
├── __init__.py
├── configuration_act_xsa.py
├── modeling_act_xsa.py
└── processor_act_xsa.py
cd lehome-challenge
python -m scripts.eval \
--policy_type lerobot \
--policy_path /path/to/downloaded/pretrained_model \
--garment_type "top_long" \
--dataset_root Datasets/example/top_long_merged \
--num_episodes 5 \
--enable_cameras \
--device cpu
可选衣服类型:top_long、top_short、pant_long、pant_short