DexForce · yuecideng · May 21, 2026 · May 18, 2026 · May 18, 2026 · May 18, 2026
diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
@@ -75,7 +75,10 @@ jobs:
       - name: Build docs
         shell: bash
         run: |
-          pip install -e . --extra-index-url http://pyp.open3dv.site:2345/simple/ --trusted-host pyp.open3dv.site
+          pip install -e ".[gensim]" \
+            --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+            --trusted-host pyp.open3dv.site \
+            --extra-index-url https://download.blender.org/pypi/
           pip install -r docs/requirements.txt
           python3 docs/scripts/sync_readme.py
           cd ${GITHUB_WORKSPACE}/docs
@@ -136,7 +139,10 @@ jobs:
       - uses: actions/checkout@v4
       - name: Run tests
         run: |
-          pip install -e . --extra-index-url http://pyp.open3dv.site:2345/simple/ --trusted-host pyp.open3dv.site
+          pip install -e ".[gensim]" \
+            --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+            --trusted-host pyp.open3dv.site \
+            --extra-index-url https://download.blender.org/pypi/
           echo "Unit test Start"
           export HF_ENDPOINT=https://hf-mirror.com
           pytest tests

diff --git a/docs/source/features/agents.md → .../source/features/generative_sim/agents.md b/docs/source/features/agents.md → .../source/features/generative_sim/agents.md
@@ -1,4 +1,4 @@
-# EmbodiAgent
+# EmbodiAgent（aborted）
 
 EmbodiAgent is a hierarchical multi-agent system that enables robots to perform complex manipulation tasks through closed-loop planning, code generation, and validation. The system combines vision-language models (VLMs) and large language models (LLMs) to translate high-level goals into executable robot actions.
 
@@ -169,7 +169,7 @@ embodichain/agents/
 
 ## See Also
 
-- [Online Data Streaming](online_data.md) — Streaming live simulation data for training
-- [RL Architecture](../overview/rl/index.rst) — RL training pipeline and algorithms
-- [Atomic Actions Tutorial](../tutorial/atomic_actions.rst) — Action primitives used by the CodeAgent
-- [Supported Tasks](../resources/task/index.rst) — Available task environments
+- [Online Data Streaming](../online_data.md) — Streaming live simulation data for training
+- [RL Architecture](../../overview/rl/index.rst) — RL training pipeline and algorithms
+- [Atomic Actions Tutorial](../../tutorial/atomic_actions.rst) — Action primitives used by the CodeAgent
+- [Supported Tasks](../../resources/task/index.rst) — Available task environments
diff --git a/docs/source/features/generative_sim/index.rst b/docs/source/features/generative_sim/index.rst
@@ -0,0 +1,9 @@
+Generative Simulation
+=====================
+
+Generative Simulation collects EmbodiChain features for generating simulation-ready assets and executing agent-driven task workflows.
+
+.. toctree::
+   :maxdepth: 2
+
+   SimReady Asset Pipeline <simready_pipeline.md>
diff --git a/docs/source/features/generative_sim/simready_pipeline.md b/docs/source/features/generative_sim/simready_pipeline.md
@@ -0,0 +1,224 @@
+# SimReady Asset Pipeline
+
+The SimReady asset pipeline converts raw mesh archives into normalized simulation assets. It ingests a source mesh, preserves or bakes visual materials, cleans mesh topology, estimates real-world scale and semantics with multimodal LLMs, and exports assets that can be loaded directly in EmbodiChain simulations.
+
+## Quick Start
+
+Run the pipeline on a single asset directory:
+
+```bash
+python -m embodichain.gen_sim.simready_pipeline.cli.start \
+    --input_dir /path/to/raw_mesh_folder \
+    --output_root /path/to/output_folder \
+    --category YourCategory
+```
+
+Preview the generated SimReady mesh:
+
+```bash
+python -m embodichain preview-asset \
+    --asset_path /path/to/sim_ready_asset_or_usd_asset \
+    --asset_type rigid
+```
+
+## Prerequisites
+
+The full pipeline uses Blender, trimesh, pyrender, and an OpenAI-compatible multimodal chat completions endpoint. Install EmbodiChain with the `gensim` extra and enable both the EmbodiChain package index and Blender package index.
+
+Install from PyPI with `uv`:
+
+```bash
+uv pip install "embodichain[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+Install from source with `uv`:
+
+```bash
+git clone https://github.com/DexForce/EmbodiChain.git
+cd EmbodiChain
+uv pip install -e ".[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+Install from PyPI with `pip`:
+
+```bash
+pip install "embodichain[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+Install from source with `pip`:
+
+```bash
+git clone https://github.com/DexForce/EmbodiChain.git
+cd EmbodiChain
+pip install -e ".[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+Set the OpenAI-compatible LLM api(OpenAI, Gemini, Doubao, etc.) before running the pipeline, or configure them in `embodichain/gen_sim/simready_pipeline/configs/gen_config.json`. Environment variables override the JSON config.
+
+OpenAI-compatible API example:
+
+```bash
+export OPENAI_API_KEY="your-openai-api-key"
+export OPENAI_MODEL="gpt-4o"
+export OPENAI_BASE_URL="https://api.openai.com/v1"
+```
+
+## Processing Flow
+
+The command above runs the full parser sequence:
+
+- **Ingest**: finds the first parseable mesh (`.glb`, `.gltf`, `.obj`, `.ply`, `.stl`), archives the raw input, and writes a canonical `asset_source/asset.obj`.
+- **Visual processing**: by default, Blender remeshes the source mesh, unwraps UVs, and bakes diffuse and normal textures. With `--simple`, ingest uses trimesh only and skips Blender remesh/bake.
+- **Inspection**: detects whether the normalized source is a mesh, articulation, or scene.
+- **Geometry processing**: cleans topology and applies Blender decimation to the canonical mesh.
+- **SimReady finalization**: renders multi-view images, uses the LLM to infer object orientation, physical dimensions, and semantics, then exports `asset_simready/asset_simready.obj`.
+- **Physics and USD export**: infers physics properties and writes a USD package when possible.
+- **Internal preview assets**: generates thumbnails and internal metadata for asset browsing.
+
+## Output Layout
+
+Each processed asset is written under a generated asset ID:
+
+```text
+simready_car/
+`-- <asset_id>/
+    |-- asset_archive/          # Raw source directory copy
+    |-- asset_source/           # Canonical normalized source mesh and textures
+    |   |-- asset.obj
+    |   |-- asset.mtl
+    |   |-- diffuse.png
+    |   `-- normal.png
+    |-- asset_simready/         # Final oriented and scaled mesh
+    |   `-- asset_simready.obj
+    |-- asset_usd/              # USD export
+    `-- asset.json              # Metadata, geometry, semantics, physics, and paths
+```
+
+Use `asset_simready/asset_simready.obj` or `asset_usd/` for simulation preview and downstream scene construction.
+
+## Command-Line Arguments
+
+| Argument | Description | Default |
+| :--- | :--- | :--- |
+| `--input_dir` | Directory containing the raw asset files. | **required** |
+| `--output_root` | Directory where processed assets are written. | **required** |
+| `--category` | Category hint passed into the pipeline, such as `car`, `bowl`, or `chair`. | **required** |
+| `--simple` | Use trimesh-only ingest and skip Blender remesh/bake during ingest. Geometry cleanup later in the pipeline still uses Blender. | `False` |
+
+## Configuration
+
+Pipeline hyperparameters live in `embodichain/gen_sim/simready_pipeline/configs/gen_config.json`. The main hyperparameters are as follow:
+
+### Mesh Processing
+
+```json
+"mesh_processing": {
+  "blender_remesh_bake": {
+    "remesh": {
+      "voxel_size": 0.01,
+      "min_voxel_size_ratio": 0.005,
+      "use_smooth_shade": true
+    },
+    "decimate": {
+      "ratio": 0.9
+    },
+    "uv": {
+      "angle_limit": 66.0,
+      "island_margin": 0.02
+    },
+    "bake": {
+      "texture_size": 2048,
+      "cage_extrusion_ratio": 0.05
+    }
+  },
+  "blender_cleanup_decimate": {
+    "enabled": true,
+    "cleanup": {
+      "merge_dist": 0.00001,
+      "remove_non_manifold": true,
+      "triangulate": false
+    },
+    "simplify": {
+      "ratio": 0.5,
+      "weld_distance": 0.0001,
+      "collapse_triangulate": true
+    }
+  },
+}
+```
+
+`blender_remesh_bake` controls the default ingest path when `--simple` is not provided. It remeshes the raw mesh, decimates it, unwraps UVs, and bakes textures.
+
+`blender_cleanup_decimate` controls the later geometry parser stage. It uses Blender mesh operators and the Blender Decimate modifier to clean and simplify the canonical mesh.
+
+
+### LLM
+
+```json
+"llm": {
+  "openai_compatible": {
+    "api_key": "",
+    "model": "gpt-4o",
+    "base_url": "https://api.openai.com/v1",
+    "default_query": {}
+  }
+}
+```
+
+This section configures the multimodal LLM used for object classification, orientation selection, dimension inference, semantic annotation, and physics inference. Any provider that supports the OpenAI-compatible chat completions API can be used by changing `api_key`, `model`, `base_url`, and optional `default_query` parameters.
+
+For Azure-style OpenAI-compatible endpoints that require an API version query parameter, use `default_query`:
+
+```json
+"llm": {
+  "openai_compatible": {
+    "api_key": "your-api-key",
+    "model": "gpt-4o",
+    "base_url": "your_api",
+    "default_query": {
+      "api-version": "2025-01-01-preview"
+    }
+  }
+}
+```
+
+## Default vs Simple Ingest
+
+The default command uses Blender during ingest:
+
+```bash
+python -m embodichain.gen_sim.simready_pipeline.cli.start \
+    --input_dir /path/to/raw_mesh_folder \
+    --output_root /path/to/output_folder \
+    --category YourCategory
+```
+
+Use `--simple` when you want faster trimesh-only ingest:
+
+```bash
+python -m embodichain.gen_sim.simready_pipeline.cli.start \
+    --input_dir /path/to/raw_mesh_folder \
+    --output_root /path/to/output_folder \
+    --category YourCategory \
+    --simple
+```
+
+The simple mode only affects the ingest step. The downstream geometry parser still uses Blender cleanup and decimation unless `mesh_processing.blender_cleanup_decimate.enabled` is set to `false`.
+
+## See Also
+
+- [Asset Preview](../interaction/preview_asset.md): Load generated meshes and USD assets in the simulator.
+- [Installation](../../quick_start/install.md): Install EmbodiChain with Blender and rendering dependencies.
+- [Toolkits](../toolkits/index.rst): Other asset preparation utilities.
diff --git a/docs/source/features/online_data.md b/docs/source/features/online_data.md
@@ -148,6 +148,6 @@ python examples/agents/datasets/online_dataset_demo.py
 
 ## See Also
 
-- [EmbodiAgent](agents.md) — Hierarchical agent that uses online data for training
+- [EmbodiAgent](generative_sim/agents.md) — Hierarchical agent that uses online data for training
 - [RL Architecture](../overview/rl/index.rst) — RL training pipeline
 - [Data Generation Tutorial](../tutorial/data_generation.rst) — Generating offline datasets
diff --git a/docs/source/guides/cli.md b/docs/source/guides/cli.md
@@ -27,6 +27,38 @@ python -m embodichain.data download --all
 
 ---
 
+## SimReady Asset Pipeline
+
+Convert a raw mesh asset directory into sim_ready assets for simulation.
+
+```bash
+# Run the full SimReady pipeline on a single asset directory
+python -m embodichain.gen_sim.simready_pipeline.cli.start \
+    --input_dir /path/to/raw_mesh_folder \
+    --output_root /path/to/output_folder \
+    --category YourCategory
+
+# Use trimesh-only ingest for source normalization
+python -m embodichain.gen_sim.simready_pipeline.cli.start \
+    --input_dir /path/to/raw_mesh_folder \
+    --output_root /path/to/output_folder \
+    --category YourCategory \
+    --simple
+```
+
+### Arguments
+
+| Argument | Default | Description |
+|---|---|---|
+| ``--input_dir`` | *(required)* | Directory containing the raw asset files |
+| ``--output_root`` | *(required)* | Directory where processed assets are written |
+| ``--category`` | *(required)* | Category hint passed into the pipeline |
+| ``--simple`` | ``False`` | Use trimesh-only ingest and skip Blender remesh/bake during ingest |
+
+The generated output contains the canonical source mesh under ``asset_source/``, the final SimReady mesh under ``asset_simready/``, and USD export files under ``asset_usd/`` when export succeeds.
+
+---
+
 ## Preview Asset
 
 Preview a USD or mesh asset in the simulation without writing code.

diff --git a/docs/source/index.rst b/docs/source/index.rst
@@ -40,7 +40,7 @@ Table of Contents
    :glob:
 
    features/online_data.md
-   features/agents.md
+   features/generative_sim/index*
    features/workspace_analyzer/index*
    features/interaction/index*
    features/toolkits/index*

diff --git a/docs/source/quick_start/install.md b/docs/source/quick_start/install.md
@@ -77,6 +77,50 @@ cd EmbodiChain
 pip install -e . --extra-index-url http://pyp.open3dv.site:2345/simple/ --trusted-host pyp.open3dv.site
 ```
 
+### Generative Simulation Dependencies
+
+If you want to use the generative simulation features, install EmbodiChain with the `gensim` extra. This installs the additional rendering and asset-processing dependencies, including `pyrender` and `bpy`. The `bpy` wheel is distributed from Blender's package index, so the Blender index must be included in the install command.
+
+**Install from PyPI with `uv`:**
+
+```bash
+uv pip install "embodichain[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+**Install from source with `uv`:**
+
+```bash
+git clone https://github.com/DexForce/EmbodiChain.git
+cd EmbodiChain
+uv pip install -e ".[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+**Install from PyPI with `pip`:**
+
+```bash
+pip install "embodichain[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
+**Install from source with `pip`:**
+
+```bash
+git clone https://github.com/DexForce/EmbodiChain.git
+cd EmbodiChain
+pip install -e ".[gensim]" \
+    --extra-index-url http://pyp.open3dv.site:2345/simple/ \
+    --trusted-host pyp.open3dv.site \
+    --extra-index-url https://download.blender.org/pypi/
+```
+
 ## Verify Installation
 
 Run the demo script to confirm everything is set up correctly: