Skip to main content

Imitation Learning Overview

DISCOVERSE provides a complete imitation learning workflow, supporting end-to-end training from data collection to policy deployment. This chapter introduces four mainstream imitation learning algorithms integrated in the framework.

đŸŽ¯ Supported Algorithms​

Currently, DISCOVERSE supports the following four imitation learning algorithms:

1. ACT (Action Chunking with Transformers)​

  • Data format: HDF5
  • Use case: Complex manipulation tasks requiring long sequence planning

2. DP (Diffusion Policy)​

  • Data format: Zarr
  • Use case: Multimodal action distributions, complex manipulation skills

3. RDT (Robotics Diffusion Transformer)​

  • Data format: HDF5
  • Use case: Multi-task learning, general robot skills

4. OpenPI (Open-source Policy Interface)​

  • Data format: HDF5
  • Use case: Rapid prototyping, few-shot learning

🔄 Workflow​

The complete imitation learning workflow includes the following steps:

1. Data Generation​

Automatically generate demonstration data, 100x more efficient than the real world

2. Data Format Conversion​

Convert to the required format for each algorithm:

  • ACT/RDT/OpenPI: Raw data īŋŊ?HDF5
  • DP: Raw data īŋŊ?Zarr