admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-26 11:16:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

docs(vlabench): align page with adding_benchmarks template

Browse Source 
Rework docs/source/vlabench.mdx to follow the standard benchmark doc
structure: intro + links + available tasks (with cleaner counts and
comma-list shortcuts) + installation + eval + recommended episodes +
policy I/O + training + reproducing results.

- Point everything at lerobot/smolvla_vlabench (the released
  checkpoint) rather than a generic "your-policy-id" placeholder.
- Document the three valid `--env.task` forms (single, comma list,
  suite shortcut).
- Drop the "Evaluation tracks" table — it reflected the upstream
  paper rather than what the LeRobot env exposes, and was confusing.
- Add an explicit "Reproducing published results" section pointing
  at the CI smoke eval.

Made-with: Cursor
This commit is contained in:
Pepijn
2026-04-17 13:45:18 +01:00
parent 893b10b825
commit 33864a32c2
1 changed files with 54 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/vlabench.mdx
+54 -60
View File
@@ -1,59 +1,49 @@
# VLABench # VLABench
VLABench is a large-scale benchmark for **language-conditioned robotic manipulation with long-horizon reasoning**. It provides 100 task categories across 2000+ objects, evaluating six dimensions of robot intelligence: mesh & texture understanding, spatial reasoning, world knowledge transfer, semantic instruction comprehension, physical law understanding, and long-horizon reasoning. VLABench is built on MuJoCo/dm_control and uses a Franka Panda 7-DOF arm. [VLABench](https://github.com/OpenMOSS/VLABench) is a large-scale benchmark for **language-conditioned robotic manipulation with long-horizon reasoning**. It covers 100 task categories across 2,000+ objects and evaluates six dimensions of robot intelligence: mesh & texture understanding, spatial reasoning, world-knowledge transfer, semantic instruction comprehension, physical-law understanding, and long-horizon planning. Built on MuJoCo / dm_control with a Franka Panda 7-DOF arm.
- Paper: [VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning](https://arxiv.org/abs/2412.18194) - Paper: [VLABench: A Large-Scale Benchmark for Language-Conditioned Robotics Manipulation with Long-Horizon Reasoning](https://arxiv.org/abs/2412.18194)
- GitHub: [OpenMOSS/VLABench](https://github.com/OpenMOSS/VLABench) - GitHub: [OpenMOSS/VLABench](https://github.com/OpenMOSS/VLABench)
- Project website: [vlabench.github.io](https://vlabench.github.io) - Project website: [vlabench.github.io](https://vlabench.github.io)
- Pretrained policy: [`lerobot/smolvla_vlabench`](https://huggingface.co/lerobot/smolvla_vlabench)
## Available tasks ## Available tasks
VLABench includes **two task suites** covering **100 task categories**: VLABench ships two task suites covering **43 task categories** in LeRobot's `--env.task` surface:
| Suite | CLI name | Tasks | Description | | Suite | CLI name | Tasks | Description |
| --------- | ----------- | ----- | ---------------------------------------------------------------- | | --------- | ----------- | ----- | ---------------------------------------------------------------- |
| Primitive | `primitive` | 21 | Single/few skill combinations (select, insert, physics QA) | | Primitive | `primitive` | 21 | Single / few-skill combinations (select, insert, physics QA) |
| Composite | `composite` | 22 | Multi-step reasoning and long-horizon planning (cook, rearrange) | | Composite | `composite` | 22 | Multi-step reasoning and long-horizon planning (cook, rearrange) |
### Primitive tasks **Primitive tasks:** `select_fruit`, `select_toy`, `select_chemistry_tube`, `add_condiment`, `select_book`, `select_painting`, `select_drink`, `insert_flower`, `select_billiards`, `select_ingredient`, `select_mahjong`, `select_poker`, and physical-reasoning tasks (`density_qa`, `friction_qa`, `magnetism_qa`, `reflection_qa`, `simple_cuestick_usage`, `simple_seesaw_usage`, `sound_speed_qa`, `thermal_expansion_qa`, `weight_qa`).
Includes `select_fruit`, `select_toy`, `select_chemistry_tube`, `add_condiment`, `select_book`, `select_painting`, `select_drink`, `insert_flower`, `select_billiards`, `select_ingredient`, `select_mahjong`, `select_poker`, and physical reasoning tasks (`density_qa`, `friction_qa`, `magnetism_qa`, `reflection_qa`, `simple_cuestick_usage`, `simple_seesaw_usage`, `sound_speed_qa`, `thermal_expansion_qa`, `weight_qa`). **Composite tasks:** `cluster_billiards`, `cluster_book`, `cluster_drink`, `cluster_toy`, `cook_dishes`, `cool_drink`, `find_unseen_object`, `get_coffee`, `hammer_nail`, `heat_food`, `make_juice`, `play_mahjong`, `play_math_game`, `play_poker`, `play_snooker`, `rearrange_book`, `rearrange_chemistry_tube`, `set_dining_table`, `set_study_table`, `store_food`, `take_chemistry_experiment`, `use_seesaw_complex`.
### Composite tasks `--env.task` accepts three forms:
Includes `cluster_billiards`, `cluster_book`, `cluster_drink`, `cluster_toy`, `cook_dishes`, `cool_drink`, `find_unseen_object`, `get_coffee`, `hammer_nail`, `heat_food`, `make_juice`, `play_mahjong`, `play_math_game`, `play_poker`, `play_snooker`, `rearrange_book`, `rearrange_chemistry_tube`, `set_dining_table`, `set_study_table`, `store_food`, `take_chemistry_experiment`, `use_seesaw_complex`. - a single task name (`select_fruit`)
- a comma-separated list (`select_fruit,heat_food`)
### Evaluation tracks - a suite shortcut (`primitive`, `composite`, or `primitive,composite`)
VLABench defines five standard evaluation tracks:
| Track | Focus |
| ----- | ----------------------------- |
| 1 | In-distribution task learning |
| 2 | Cross-category generalization |
| 3 | Commonsense reasoning |
| 4 | Semantic instruction |
| 6 | Unseen texture robustness |
## Installation ## Installation
After following the LeRobot installation instructions: After following the standard LeRobot installation instructions:
```bash ```bash
pip install -e ".[vlabench]" pip install -e ".[vlabench]"
``` ```
VLABench also requires downloading simulation assets: VLABench also ships simulation assets separately. Clone the repo and download them:
```bash ```bash
# Clone the VLABench repo and download assets
git clone https://github.com/OpenMOSS/VLABench.git git clone https://github.com/OpenMOSS/VLABench.git
cd VLABench cd VLABench
python scripts/download_assets.py python scripts/download_assets.py
``` ```
<Tip> <Tip>
VLABench requires Linux (`sys_platform == 'linux'`) and Python 3.10+. Set the MuJoCo rendering backend before training or evaluation: VLABench requires Linux (`sys_platform == 'linux'`) and Python 3.10+. Set the MuJoCo rendering backend before running:
```bash ```bash
export MUJOCO_GL=egl # for headless servers (HPC, cloud) export MUJOCO_GL=egl # for headless servers (HPC, cloud)
@@ -63,26 +53,37 @@ export MUJOCO_GL=egl # for headless servers (HPC, cloud)
## Evaluation ## Evaluation
### Default evaluation (recommended) ### Single-task evaluation (recommended for quick iteration)
Evaluate on a single task (10 episodes):
```bash ```bash
lerobot-eval \ lerobot-eval \
--policy.path="your-policy-id" \ --policy.path=lerobot/smolvla_vlabench \
--env.type=vlabench \ --env.type=vlabench \
--env.task=select_fruit \ --env.task=select_fruit \
--eval.batch_size=1 \ --eval.batch_size=1 \
--eval.n_episodes=10 --eval.n_episodes=10
``` ```
### Suite-wide evaluation ### Multi-task evaluation
Evaluate across all primitive tasks: Pass a comma-separated list of tasks:
```bash ```bash
lerobot-eval \ lerobot-eval \
--policy.path="your-policy-id" \ --policy.path=lerobot/smolvla_vlabench \
--env.type=vlabench \
--env.task=select_fruit,select_toy,add_condiment,heat_food \
--eval.batch_size=1 \
--eval.n_episodes=10
```
### Suite-wide evaluation
Run an entire suite (all 21 primitives or all 22 composites):
```bash
lerobot-eval \
--policy.path=lerobot/smolvla_vlabench \
--env.type=vlabench \ --env.type=vlabench \
--env.task=primitive \ --env.task=primitive \
--eval.batch_size=1 \ --eval.batch_size=1 \
@@ -90,13 +91,11 @@ lerobot-eval \
--env.max_parallel_tasks=1 --env.max_parallel_tasks=1
``` ```
### Multi-suite evaluation Or both suites:
Evaluate across both primitive and composite suites:
```bash ```bash
lerobot-eval \ lerobot-eval \
--policy.path="your-policy-id" \ --policy.path=lerobot/smolvla_vlabench \
--env.type=vlabench \ --env.type=vlabench \
--env.task=primitive,composite \ --env.task=primitive,composite \
--eval.batch_size=1 \ --eval.batch_size=1 \
@@ -104,59 +103,54 @@ lerobot-eval \
--env.max_parallel_tasks=1 --env.max_parallel_tasks=1
``` ```
### Individual task evaluation ### Recommended evaluation episodes
Evaluate on specific tasks by name: **10 episodes per task** for reproducible benchmarking (210 total for the full primitive suite, 220 for composite). Matches the protocol in the VLABench paper.
```bash
lerobot-eval \
--policy.path="your-policy-id" \
--env.type=vlabench \
--env.task=select_fruit,heat_food \
--eval.batch_size=1 \
--eval.n_episodes=10
```
## Policy inputs and outputs ## Policy inputs and outputs
**Observations:** **Observations:**
- `observation.state` — 7-dim end-effector state (position xyz, euler xyz, gripper) - `observation.state` — 7-dim end-effector state (position xyz + Euler xyz + gripper)
- `observation.images.image` — front camera view, 480x480 HWC uint8 - `observation.images.image` — front camera, 480×480 HWC uint8
- `observation.images.second_image` — second camera view, 480x480 HWC uint8 - `observation.images.second_image` — second camera, 480×480 HWC uint8
- `observation.images.wrist_image` — wrist camera view, 480x480 HWC uint8 - `observation.images.wrist_image` — wrist camera, 480×480 HWC uint8
**Actions:** **Actions:**
- Continuous control in `Box(-1, 1, shape=(7,))` — 3D position + 3D euler orientation + 1D gripper - Continuous control in `Box(-1, 1, shape=(7,))` — 3D position + 3D Euler orientation + 1D gripper.
### Recommended evaluation episodes
For reproducible benchmarking, use **10 episodes per task**. For the full primitive suite this gives 210 episodes; for the full composite suite, 220 episodes.
## Training ## Training
### Datasets ### Datasets
Pre-collected VLABench datasets in LeRobot format are available on the Hugging Face Hub: Pre-collected VLABench datasets in LeRobot format on the Hub:
- Primitive tasks: [VLABench/vlabench_primitive_ft_lerobot_video](https://huggingface.co/datasets/VLABench/vlabench_primitive_ft_lerobot_video) (5,000 episodes, 128 tasks, 480x480 images) - [`VLABench/vlabench_primitive_ft_lerobot_video`](https://huggingface.co/datasets/VLABench/vlabench_primitive_ft_lerobot_video) — 5,000 episodes, 128 tasks, 480×480 images.
- Composite tasks: [VLABench/vlabench_composite_ft_lerobot_video](https://huggingface.co/datasets/VLABench/vlabench_composite_ft_lerobot_video) (5,977 episodes, 167 tasks, 224x224 images) - [`VLABench/vlabench_composite_ft_lerobot_video`](https://huggingface.co/datasets/VLABench/vlabench_composite_ft_lerobot_video) — 5,977 episodes, 167 tasks, 224×224 images.
### Example training command ### Example training command
Fine-tune a SmolVLA base on the primitive suite:
```bash ```bash
lerobot-train \ lerobot-train \
--policy.type=smolvla \ --policy.type=smolvla \
--policy.repo_id=${HF_USER}/vlabench-test \ --policy.repo_id=${HF_USER}/smolvla_vlabench_primitive \
--policy.load_vlm_weights=true \ --policy.load_vlm_weights=true \
--policy.push_to_hub=true \
--dataset.repo_id=VLABench/vlabench_primitive_ft_lerobot_video \ --dataset.repo_id=VLABench/vlabench_primitive_ft_lerobot_video \
--env.type=vlabench \ --env.type=vlabench \
--env.task=select_fruit \ --env.task=select_fruit \
--output_dir=./outputs/ \ --output_dir=./outputs/smolvla_vlabench_primitive \
--steps=100000 \ --steps=100000 \
--batch_size=4 \ --batch_size=4 \
--eval_freq=5000 \
--eval.batch_size=1 \ --eval.batch_size=1 \
--eval.n_episodes=1 \ --eval.n_episodes=1 \
--eval_freq=1000 --save_freq=10000
``` ```
## Reproducing published results
The released checkpoint [`lerobot/smolvla_vlabench`](https://huggingface.co/lerobot/smolvla_vlabench) was trained on the primitive-suite dataset above and is evaluated with the [Single-task](#single-task-evaluation-recommended-for-quick-iteration) / [Suite-wide](#suite-wide-evaluation) commands. CI runs a 10-primitive-task smoke eval (one episode each) on every PR touching the benchmark.