Tutorials

This section is designed for developers who want to gain a deeper understanding of and extend FluxVLA. The tutorials are organized in a “big picture first, then modules, then scenarios” order to help you quickly build a comprehensive understanding from configuration to training and inference deployment.

Recommended reading path:

1. Start with the overall architecture: Understand the code layering and main execution chain.

2. Then explore the core modules: Master the extension methods for configuration, models, modules, engines, and data configuration.

3. Finally, review inference in practice: Complete hands-on deployment with specific robot platforms.

Tutorial Contents

• FluxVLA Code Architecture Overview: FluxVLA code architecture overview and reading entry point.

• Configuration: Configuration system in detail (model, data, training, evaluation).

• Adding Custom Models: How to add a custom model.

• Adding Custom Modules: How to add a custom module and register it with the framework.

• Adding Custom Engines: How to extend training/evaluation engines.

• Training and Deployment with Private Datasets: Dataset and data configuration extension guide.

• Inference Deployment: Inference tutorials for different robot platforms.

• FluxVLA Code Architecture Overview
  • 1. Code Organization Overview
  • 2. End-to-End Data and Training Flow
  • 3. Typical Call Chain
  • 4. Recommended Reading Order
• Configuration
  • Model Configuration
  • Data Configuration
  • Training Configuration
  • Inference / Evaluation Configuration
• Adding Custom Models
  • Overview
  • Architecture Overview
  • Adding a New VLA Model
  • Adding a New VLM Backbone
  • Adding a New VLA Head
  • Adding a New Projector
  • Composing Models in Configuration Files
  • Complete Examples: Using Existing Configurations as References
  • Core Checklist
  • Frequently Asked Questions
• Adding Custom Modules
  • Overview
  • Adding a New VLA Head
  • Core Checklist
  • Frequently Asked Questions
• Adding Custom Engines
  • Overview
  • Architecture Overview
  • Adding a New Training Runner
  • Adding a New Inference Runner
  • Adding a New Operator
  • Using the New Engine in Configuration Files
  • Complete Example: Adapting a New Robot Using Aloha as a Reference
  • Adding a New Evaluation Runner (Optional)
  • Core Checklist
  • Frequently Asked Questions
• Training and Deployment with Private Datasets
  • Overview
  • Overall Workflow
  • Step 1: Verify the Dataset Structure
  • Step 2: Configure the Model (model)
  • Step 3: Configure the Training Data (train_dataloader)
  • Step 4: Configure Training Parameters (runner)
  • Step 5: Configure Inference Deployment (inference)
  • Step 6: Launch Training
  • Step 7: Real-Robot Deployment
  • Complete Example: Aloha Dual-Arm Robot
  • Frequently Asked Questions
• Inference Deployment
  • Aloha Physical Deployment
  • Tron2 Physical Deployment
  • UR3 Physical Deployment
  • Inference Acceleration
  • RTC (Real-Time Chunking)