Views
No views yet
[TODO: insert HuggingFace repo link or Docker image URI]1git clone https://github.com/lehome-official/lehome-challenge.git
2
3cd lehome-challenge
4uv sync
5
6cd third_party
7git clone https://github.com/lehome-official/IsaacLab.git
8cd ..
9
10source .venv/bin/activate
11./third_party/IsaacLab/isaaclab.sh -i none
12
13uv pip install -e ./source/lehome1# Simulation assets (scenes, objects, robots)
2hf download lehome/asset_challenge --repo-type dataset --local-dir Assets
3
4# Dataset metadata (required for LeRobot policy loading)
5hf download lehome/dataset_challenge_merged --repo-type dataset --local-dir Datasets/exampleoutputs/train/smolvla_four_merge_3/checkpoints/1# Top Long
2PYNPUT_BACKEND=dummy xvfb-run -a python -m scripts.eval \
3 --policy_type lerobot \
4 --policy_path outputs/train/smolvla_four_merge_3/checkpoints/last/pretrained_model \
5 --garment_type "top_long" \
6 --dataset_root Datasets/example/four_types_merged \
7 --num_episodes 5 \
8 --enable_cameras \
9 --device cpu \
10 --headless \
11 --task_description "fold the garment on the table"
12
13# Top Short
14PYNPUT_BACKEND=dummy xvfb-run -a python -m scripts.eval \
15 --policy_type lerobot \
16 --policy_path outputs/train/smolvla_four_merge_3/checkpoints/last/pretrained_model \
17 --garment_type "top_short" \
18 --dataset_root Datasets/example/four_types_merged \
19 --num_episodes 5 \
20 --enable_cameras \
21 --device cpu \
22 --headless \
23 --task_description "fold the garment on the table"
24
25# Pant Long
26PYNPUT_BACKEND=dummy xvfb-run -a python -m scripts.eval \
27 --policy_type lerobot \
28 --policy_path outputs/train/smolvla_four_merge_3/checkpoints/last/pretrained_model \
29 --garment_type "pant_long" \
30 --dataset_root Datasets/example/four_types_merged \
31 --num_episodes 5 \
32 --enable_cameras \
33 --device cpu \
34 --headless \
35 --task_description "fold the garment on the table"
36
37# Pant Short
38PYNPUT_BACKEND=dummy xvfb-run -a python -m scripts.eval \
39 --policy_type lerobot \
40 --policy_path outputs/train/smolvla_four_merge_3/checkpoints/last/pretrained_model \
41 --garment_type "pant_short" \
42 --dataset_root Datasets/example/four_types_merged \
43 --num_episodes 5 \
44 --enable_cameras \
45 --device cpu \
46 --headless \
47 --task_description "fold the garment on the table"| Parameter | Value | Notes |
|---|---|---|
--policy_type | lerobot | Use LeRobot policy wrapper |
--dataset_root | Datasets/example/four_types_merged | Required for metadata loading |
--device | cpu | Simulation only supports CPU |
--task_description | "fold the garment on the table" | Unified prompt for all garment types |
--headless | flag | Use with xvfb-run -a on headless servers |
PYNPUT_BACKEND=dummy xvfb-run -a python -m scripts.eval ...rollout_results.txt for detailed per-garment results.| Category | Success Rate |
|---|---|
| Top Long | 63.33% |
| Top Short | 38.33% |
| Pant Long | 45% |
| Pant Short | 81.67% |
| Average | 57.082% |
uv pip install --upgrade --no-cache-dir --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1 Compile with TORCH_USE_CUDA_DSA to enable device-side assertions
NVIDIA GeForce RTX 5090 with CUDA capability sm_120 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90. If you want to use the NVIDIA GeForce RTX 5090 GPU with PyTorch, please check the instructions at Start Locally | PyTorch