mirror of
https://github.com/huggingface/lerobot.git
synced 2026-07-26 11:16:00 +00:00
Pepijn
63dbdf0e55
- The nightly base image has stale lerobot code (pre-dates VLABenchEnv addition), so COPY . . doesn't propagate into .venv. Run `uv pip install --no-deps -e .` after COPY, same fix as robocasa365. - Coerce ee_state to exactly shape (7,) so SyncVectorEnv.concatenate doesn't raise "Output array is the wrong shape" when VLABench's actual ee_state length differs from our declared observation_space. Made-with: Cursor
58 lines
2.8 KiB
Docker
58 lines
2.8 KiB
Docker
# Copyright 2025 The HuggingFace Inc. team. All rights reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
|
|
# Benchmark image for VLABench integration tests.
|
|
# Extends the nightly GPU image with the PR's source code and VLABench setup.
|
|
#
|
|
# Build: docker build -f docker/Dockerfile.benchmark.vlabench -t lerobot-benchmark-vlabench .
|
|
# Run: docker run --gpus all --rm lerobot-benchmark-vlabench lerobot-eval ...
|
|
|
|
FROM huggingface/lerobot-gpu:latest
|
|
|
|
# Install VLABench from GitHub (not on PyPI) and pin MuJoCo/dm-control.
|
|
# Shallow-clone without submodule recursion (nested SSH-only submodules fail in CI).
|
|
# Editable install (-e) because VLABench/utils/ has no __init__.py, so
|
|
# find_packages() omits it from wheels; editable mode uses the source tree directly.
|
|
# rrt-algorithms has the same packaging issue (rrt/ dir missing __init__.py).
|
|
# Patch: constant.py calls os.listdir on ~100 asset/obj/meshes/* dirs at import
|
|
# time. Guard the call so missing dirs return [] instead of crashing (in case
|
|
# the asset download is partial).
|
|
RUN git clone --depth 1 https://github.com/OpenMOSS/VLABench.git ~/VLABench && \
|
|
git clone --depth 1 https://github.com/motion-planning/rrt-algorithms.git ~/rrt-algorithms && \
|
|
python3 -c "\
|
|
import pathlib; \
|
|
p = pathlib.Path.home() / 'VLABench/VLABench/configs/constant.py'; \
|
|
t = p.read_text(); \
|
|
p.write_text(t.replace( \
|
|
'subdirs = os.listdir(xml_dir)', \
|
|
'if not os.path.isdir(xml_dir): return []\n subdirs = os.listdir(xml_dir)'))" && \
|
|
uv pip install --no-cache -e ~/VLABench -e ~/rrt-algorithms \
|
|
mujoco==3.2.2 dm-control==1.0.22 \
|
|
open3d colorlog scikit-learn openai gdown
|
|
|
|
# Download VLABench mesh assets from Google Drive. Task configs reference
|
|
# object meshes (obj/meshes/fruit/, containers/basket/, etc.); without them
|
|
# the task builder crashes with IndexError on random.choice([]).
|
|
# gdown handles Google Drive's confirmation tokens for large files.
|
|
RUN python ~/VLABench/scripts/download_assets.py --choice all
|
|
|
|
# Overlay the PR's source code on top of the nightly image.
|
|
COPY --chown=user_lerobot:user_lerobot . .
|
|
|
|
# Re-install lerobot editably so the new source (with VLABenchEnv registration
|
|
# and updated obs handling) replaces the stale package baked into the nightly image.
|
|
RUN uv pip install --no-cache --no-deps -e .
|
|
|
|
CMD ["/bin/bash"]
|