Example Workflows#

This page provides an overview of the complete mindmap workflow and the available shortcuts to get you started quickly.

Workflow Overview#

../_images/mindmap_workflows.png

mindmap Workflow Diagram#

Complete Workflow#

To train and evaluate a mindmap model on your own custom task, follow these steps:

  1. Create a mindmap task - Define your custom robotic task

  2. Record demonstrations - Collect expert demonstrations for imitation learning

  3. Generate a mindmap dataset - Process demonstrations into mindmap format

  4. Train a mindmap model - Train the spatial memory model

  5. Evaluate open loop - Test model predictions on recorded data

  6. Evaluate closed loop - Test model performance in Isaac Lab simulation

Quick Start Options#

We provide several shortcuts to help you get started immediately:

Predefined Tasks:

Four benchmark tasks (Cube Stacking, Mug in Drawer, Drill in Box, Stick in Bin) with complete data and models ready to use.

Downloadable Resources:

  • HDF5 demonstration files for all tasks

  • Pre-generated datasets in mindmap format

  • Trained model checkpoints ready for evaluation

These resources allow you to:

  • Skip directly to any workflow step

  • Use them as references when creating your own tasks

  • Quickly test mindmap capabilities without going through the complete workflow

Getting Started#

See the Workflow Diagram above for a visual overview of the complete workflow and available shortcuts.

For detailed information about: