README.md

---
pretty_name: AssemblyWorldBench
license: other
license_name: per-source-terms
license_link: LICENSE.md
tags:
  - 3d
  - assembly
  - agents
  - benchmark
  - webmcp
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: test
        path: index/samples.parquet
extra_gated_prompt: >-
  AssemblyWorldBench embeds part geometry, manual pages and reference images from
  PartNet (via Manual-PA), IKEA-Manual, AssemblyBench (Fusion 360 Gallery) and Fantastic
  Breaks. These assets keep their original rights. Access is granted for non-commercial
  research only, and you must comply with the terms of every source dataset.
extra_gated_fields:
  Name: text
  Affiliation: text
  I will use this dataset for non-commercial research only: checkbox
  I will comply with the terms of each source dataset listed in the card: checkbox
---

# AssemblyWorldBench

AssemblyWorldBench is the benchmark of *AssemblyWorld: Rethinking 3D Assembly with
General-Purpose Agents*. An agent receives the rigid parts of one object, scattered in an
interactive 3D scene, and must assemble them by inspecting rendered views and manipulating
parts through the environment's WebMCP tools, optionally guided by a reference image or a
full assembly manual. The finished scene is scored geometrically against the ground-truth
assembly.

The benchmark has **100 evaluations over 80 shapes** from four sources, organized into five
blocks of 20. Agent results for the systems in the paper are released separately in
[`AssemblyWorld/AssemblyWorldBench-Results`](https://huggingface.co/datasets/AssemblyWorld/AssemblyWorldBench-Results).

| Block | Source dataset | Reference given to the agent | Objects | Parts per object |
|---|---|---|---|---|
| `partnet-none` | PartNet (Manual-PA split) | none | 7 chairs, 7 tables, 6 storage | 5–15 |
| `partnet-final-image` | PartNet (Manual-PA split) | final assembly image | same 20 objects | 5–15 |
| `ikea-manualbook` | IKEA-Manual | full manual, original page order | 20 furniture objects | 3–19 |
| `assemblybench-manualbook` | AssemblyBench | full step diagrams, in order | 20 industrial objects | 3–18 |
| `fantastic-breaks-none` | Fantastic Breaks | none | 20 broken objects | 2 |

Objects were selected from task properties only (category and part-count bands, fixed seed),
never from model results. The selection inputs are in `selection/`.

## Layout

```
benchmark.json                     frozen definition: blocks, samples, protocol, aggregation, provenance
launch-commands.txt                one run command per block
index/samples.parquet              one row per evaluation (block, source, sample, category, band, parts, paths)
selection/                         spec, manifest, exclusions, exclusion candidates, overrides
<block>/task.txt                   the task text of the block
<block>/<dataset>/<config_id>/
    config.json                    preparation identity and per-sample initial episode records
    <sample>.episode.zip           initial episode: the only input an agent receives
    cache/<sample>/evaluation.json ground truth for scoring: part point clouds, GT poses, equivalence groups
    cache/<sample>/reference/<mode>/  reference pages attached to the prompt (final-image, manualbook)
```

An initial episode is a [3DWebAgent](https://github.com/AssemblyWorld/3DWebAgent) episode
archive (`3dwebagent-episode` v1, runtime contract `3dwebagent-runtime-1`): MuJoCo world,
part meshes and the scattered initial state. The `cache/` directories hold ground truth and
must never be given to an agent; the runner reads only the reference pages from them, and
the evaluator reads `evaluation.json`.

## Running and scoring

The environment is served at `https://assemblyworld.github.io/3DWebAgent/`. The public
runner and official scorer are [`assembly-world-bench`](https://github.com/AssemblyWorld/assembly-world-bench)
(v1.0.0, MIT software; macOS and Linux). Install that repository with
`uv sync --locked --extra browser`, and install/authenticate the selected agent CLI and
the pinned browser MCP dependency as described in its README.

```sh
# Necessary files are downloaded from the tool's fixed HF revision automatically.
uv run assembly-world-bench download --block ikea-manualbook
uv run assembly-world-bench run --all --agent codex --model MODEL --headless
uv run assembly-world-bench status logs/RUN_ID
uv run assembly-world-bench resume logs/RUN_ID
uv run assembly-world-bench eval logs/RUN_ID
# For an existing v1 snapshot, add --benchmark /absolute/path/to/package.
# Explicit local mode never fetches or reconstructs missing data.
```

See `launch-commands.txt` for individual block commands. Task text, reference conditions,
sample identities, GT clouds, GT poses, equivalence groups and scales are frozen. GT is
never provided to the agent or served over HTTP. Only this benchmark repository is
consumed by the tool; the separate paper-results repository is a reference link.

Scoring uses protocol **`assembly-evaluation-v1`**. This release only renames the previous
`assembly-evaluation-v2` protocol marker; it does not change any numerical rule, sample,
task, initial episode, reference image or GT value. Older revisions retain the original
marker for historical use; the v1 tool accepts only the migrated package.

The final assembly is aligned to the ground truth with one global SE(3) transform, and
parts are matched within the cached geometry equivalence groups. **SCD** is the whole-shape
Chamfer distance (sum of bidirectional mean squared nearest-neighbor distances, ×1000);
**PA** is the fraction of parts whose matched Chamfer distance is at most 0.01;
**SR** is 1 when every part is correct. Alignment and matching rules are fixed.

An attempted evaluation without valid scored geometry counts PA = SR = 0, including
missing or invalid final exports. Agent-reported status never replaces a valid geometric
score; refusals remain in the evaluation denominator. SCD averages only successfully
scored samples. Evaluation
follows explicit resume chains and uses each sample's last actual attempt; it does not
fall back to earlier successful exports. Unrelated runs with duplicate samples are errors.
Partial reports show selected sample/block scores and coverage. Official Overall is emitted
only when all five blocks have final attempt records for all 100 evaluations: average the
blocks within each source, then average the four sources equally. Resource statistics
include only the selected attempt; unknown values remain null and costs are not estimated.

## Sources and terms

Every block pins a source dataset revision in `benchmark.json`. The initial episodes embed the
source part geometry, and the reference pages are the source images. These assets keep their
original rights; see `LICENSE.md` and the source cards.

| Source | Hugging Face dataset | Terms of the underlying assets |
|---|---|---|
| PartNet (Manual-PA) | [`AssemblyWorld/partnet-manualpa`](https://huggingface.co/datasets/AssemblyWorld/partnet-manualpa) | PartNet and ImagePA / Manual-PA source rights; no blanket redistribution grant verified |
| IKEA-Manual | [`AssemblyWorld/ikea-manual`](https://huggingface.co/datasets/AssemblyWorld/ikea-manual) | annotations CC BY-NC-SA 4.0; manuals and meshes keep their creators' rights |
| AssemblyBench | [`AssemblyWorld/assemblybench`](https://huggingface.co/datasets/AssemblyWorld/assemblybench) | modifications CC BY-SA 4.0 (MERL); Fusion 360 Gallery assets non-commercial research only under the Fusion 360 Gallery Dataset License |
| Fantastic Breaks | [`AssemblyWorld/fantastic-breaks`](https://huggingface.co/datasets/AssemblyWorld/fantastic-breaks) | no dataset license specified; source assets keep their original rights |

Please also cite the source datasets: Mo et al., *PartNet*, 2019; Zhang et al.,
*Manual-PA*, 2025; Wang et al., *IKEA-Manual*, 2022; Li et al., *AssemblyBench*,
2026; Lamb et al., *Fantastic Breaks*, 2023.