Dataset Preparation

The following sections describe the dataset preparation workflow for imitation learning, which consists of three main stages: Prerequisites, Recording, and Visualization.

• The Prerequisites stage outlines the necessary setup steps before you begin recording your dataset.
• The Recording stage explains how to collect data using the Web UI.
  Two recording modes are available. The table below summarizes the key differences between these modes:

	Single Task Mode	Multi-Task Mode
Purpose	Focused on collecting data for a single specific task	Designed for collecting data across multiple tasks
Environment Reset	Allows time to reset the environment between recordings	No reset time; ideal for continuous, segmented motions or diverse objects
Recording Constraints	Requires a predefined number of episodes and recording duration	No predefined limits; record freely until complete
Task Structure	Only one task is recorded per session	Multiple tasks are recorded in a single session, grouped into a scenario that tracks how many scenarios have been captured

You will learn more about how to use each mode in the Recording section.

• The Visualization stage focuses on verifying data quality by allowing users to inspect the recorded dataset.