# internrobotics/internw0-delta architecture

> Architecture diagram of the GitHub repository internrobotics/internw0-delta, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-10-02). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/internrobotics/internw0-delta
- Repository: https://github.com/internrobotics/internw0-delta

## Overview

InternW0-Δ is a robot world–action model trained on robot and human demonstrations, then used to generate actions for benchmark simulators or real robots. Its primary workflow runs from dataset preparation through pretraining and task-specific post-training to online policy inference and evaluation or deployment. The model couples video and action experts with visual memory and language conditioning; 4D teacher supervision is training-only. The README documents the broad product stages; sampled source and the SOURCE INDEX clarify selected internal relationships, while unsampled implementation details are not inferred.

Main entry points: [`src/wam/inference/online_action_policy.py`](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/inference/online_action_policy.py) (Online action policy), [`src/wam/trainer.py`](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/trainer.py) (Training loop), [`src/wam/model/world_action_model.py`](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/world_action_model.py) (World-action model), [`src/wam/datasets`](https://github.com/internrobotics/internw0-delta/tree/main/src/wam/datasets) (Dataset loading)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_data["Data preparation"]
  node_data_pipeline["Dataset loading"]
  node_video_dataset["Video dataset"]
  node_pretrain_loader["Pretraining loader"]
  node_cache[("Artifact cache<br/>[manager.py]")]
end

subgraph group_learning["Model and training"]
  node_model["World-action model"]
  node_video_expert["Video expert<br/>[wan_video_dit.py]"]
  node_action_expert["Action expert<br/>[action_dit.py]"]
  node_vlm["Language understanding<br/>[qwen_vl_encoder.py]"]
  node_distillation["4D distillation<br/>[teacher.py]"]
  node_trainer["Training loop<br/>[trainer.py]"]
  node_checkpoint["Model checkpoint<br/>[checkpoint.py]"]
end

subgraph group_runtime["Action inference"]
  node_online_policy["Online action policy"]
end

subgraph group_evaluation["Evaluation"]
  node_libero_eval["LIBERO evaluation"]
  node_libero_plus["LIBERO Plus<br/>[server.py]"]
  node_robotwin["RoboTwin evaluation<br/>[policy.py]"]
  node_robodojo["RoboDojo evaluation<br/>[server.py]"]
end

subgraph group_deployment["Robot deployment"]
  node_deployment_policy["Deployment policy<br/>[policy.py]"]
  node_controller["Robot controller<br/>[controller.py]"]
end

node_demonstrations(("Robot demonstrations"))
node_robot(("Real robot"))

node_demonstrations -->|"prepare data"| node_data_pipeline
node_data_pipeline -->|"load sources"| node_pretrain_loader
node_data_pipeline -.->|"prepare samples"| node_video_dataset
node_pretrain_loader -->|"train from data"| node_trainer
node_trainer -->|"optimize model"| node_model
node_model -->|"includes expert"| node_video_expert
node_model -->|"includes expert"| node_action_expert
node_video_expert -->|"exchange features"| node_action_expert
node_model -->|"uses conditioning"| node_vlm
node_distillation -->|"supervise training"| node_trainer
node_trainer -->|"save weights"| node_checkpoint
node_checkpoint -.->|"load weights"| node_online_policy
node_online_policy -->|"predict actions"| node_libero_eval
node_online_policy -->|"predict actions"| node_robotwin
node_online_policy -->|"serve policy"| node_libero_plus
node_online_policy -.->|"serve policy"| node_robodojo
node_robot -.->|"provide observations"| node_deployment_policy
node_deployment_policy -->|"invoke policy"| node_online_policy
node_deployment_policy -.->|"send actions"| node_controller
node_controller -.->|"control robot"| node_robot
node_cache -->|"cache artifacts"| node_trainer

click node_data_pipeline "https://github.com/internrobotics/internw0-delta/tree/main/src/wam/datasets"
click node_video_dataset "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/datasets/lerobot/robot_video_dataset.py"
click node_pretrain_loader "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/datasets/pretrain_lerobot_loader.py"
click node_cache "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/cache/manager.py"
click node_model "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/world_action_model.py"
click node_video_expert "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/backbones/wan22/wan_video_dit.py"
click node_action_expert "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/modules/experts/action_dit.py"
click node_vlm "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/modules/understanding/qwen_vl_encoder.py"
click node_distillation "https://github.com/internrobotics/internw0-delta/blob/main/tools/4D_distillation/teacher.py"
click node_trainer "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/trainer.py"
click node_checkpoint "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/training/checkpoint.py"
click node_online_policy "https://github.com/internrobotics/internw0-delta/blob/main/src/wam/inference/online_action_policy.py"
click node_libero_eval "https://github.com/internrobotics/internw0-delta/blob/main/eval/libero/eval_libero_single.py"
click node_libero_plus "https://github.com/internrobotics/internw0-delta/blob/main/eval/libero_plus/server.py"
click node_robotwin "https://github.com/internrobotics/internw0-delta/blob/main/eval/robotwin/policy.py"
click node_robodojo "https://github.com/internrobotics/internw0-delta/blob/main/eval/robodojo/server.py"
click node_deployment_policy "https://github.com/internrobotics/internw0-delta/blob/main/deploy/policy.py"
click node_controller "https://github.com/internrobotics/internw0-delta/blob/main/deploy/controller.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_data_pipeline,node_video_dataset,node_pretrain_loader,node_cache toneBlue
class node_model,node_video_expert,node_action_expert,node_vlm,node_distillation,node_trainer,node_checkpoint toneAmber
class node_online_policy toneMint
class node_libero_eval,node_libero_plus,node_robotwin,node_robodojo toneRose
class node_deployment_policy,node_controller,node_demonstrations,node_robot toneIndigo
```

## Components

### Data preparation

- [Dataset loading](https://github.com/internrobotics/internw0-delta/tree/main/src/wam/datasets): `src/wam/datasets`
- [Video dataset](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/datasets/lerobot/robot_video_dataset.py): `src/wam/datasets/lerobot/robot_video_dataset.py`
- [Pretraining loader](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/datasets/pretrain_lerobot_loader.py): `src/wam/datasets/pretrain_lerobot_loader.py`
- [Artifact cache](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/cache/manager.py): `src/wam/cache/manager.py`

### Model and training

- [World-action model](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/world_action_model.py): `src/wam/model/world_action_model.py`
- [Video expert](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/backbones/wan22/wan_video_dit.py): `src/wam/model/backbones/wan22/wan_video_dit.py`
- [Action expert](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/modules/experts/action_dit.py): `src/wam/model/modules/experts/action_dit.py`
- [Language understanding](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/model/modules/understanding/qwen_vl_encoder.py): `src/wam/model/modules/understanding/qwen_vl_encoder.py`
- [4D distillation](https://github.com/internrobotics/internw0-delta/blob/main/tools/4D_distillation/teacher.py): `tools/4D_distillation/teacher.py`
- [Training loop](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/trainer.py): `src/wam/trainer.py`
- [Model checkpoint](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/training/checkpoint.py): `src/wam/training/checkpoint.py`

### Action inference

- [Online action policy](https://github.com/internrobotics/internw0-delta/blob/main/src/wam/inference/online_action_policy.py): `src/wam/inference/online_action_policy.py`

### Evaluation

- [LIBERO evaluation](https://github.com/internrobotics/internw0-delta/blob/main/eval/libero/eval_libero_single.py): `eval/libero/eval_libero_single.py`
- [LIBERO Plus](https://github.com/internrobotics/internw0-delta/blob/main/eval/libero_plus/server.py): `eval/libero_plus/server.py`
- [RoboTwin evaluation](https://github.com/internrobotics/internw0-delta/blob/main/eval/robotwin/policy.py): `eval/robotwin/policy.py`
- [RoboDojo evaluation](https://github.com/internrobotics/internw0-delta/blob/main/eval/robodojo/server.py): `eval/robodojo/server.py`

### Robot deployment

- [Deployment policy](https://github.com/internrobotics/internw0-delta/blob/main/deploy/policy.py): `deploy/policy.py`
- [Robot controller](https://github.com/internrobotics/internw0-delta/blob/main/deploy/controller.py): `deploy/controller.py`

### Other components

- **Robot demonstrations**
- **Real robot**

## Connections

- Robot demonstrations → Dataset loading: prepare data (evidence: `README.md`)
- Dataset loading → Pretraining loader: load sources (evidence: `README.md`)
- Dataset loading → Video dataset: prepare samples (evidence: `src/wam/datasets/lerobot/robot_video_dataset.py`)
- Pretraining loader → Training loop: train from data (evidence: `src/wam/trainer.py`)
- Training loop → World-action model: optimize model (evidence: `src/wam/trainer.py`)
- World-action model → Video expert: includes expert (evidence: `README.md`)
- World-action model → Action expert: includes expert (evidence: `README.md`)
- Video expert → Action expert: exchange features (evidence: `README.md`)
- World-action model → Language understanding: uses conditioning (evidence: `README.md`)
- 4D distillation → Training loop: supervise training (evidence: `README.md`)
- Training loop → Model checkpoint: save weights (evidence: `src/wam/trainer.py`)
- Model checkpoint → Online action policy: load weights
- Online action policy → LIBERO evaluation: predict actions (evidence: `eval/libero/eval_libero_single.py`)
- Online action policy → RoboTwin evaluation: predict actions (evidence: `eval/robotwin/policy.py`)
- Online action policy → LIBERO Plus: serve policy (evidence: `eval/libero_plus/server.py`)
- Online action policy → RoboDojo evaluation: serve policy
- Real robot → Deployment policy: provide observations
- Deployment policy → Online action policy: invoke policy (evidence: `deploy/policy.py`)
- Deployment policy → Robot controller: send actions
- Robot controller → Real robot: control robot
- Artifact cache → Training loop: cache artifacts (evidence: `src/wam/trainer.py`)
