How to Take This Course#

Learning Strategies#

There are several ways to work through this learning path, depending on your goals, how much time you want to invest, and how much of a challenge you want to take on.

Option 1: As-Is#

If you’re not sure, just take the course as-is!

Buy the workspace materials from the Bill of Materials, use our pre-trained checkpoints and pre-collected datasets, and follow the course as-is.

Tip

This is the fastest way to experience the full sim-to-real workflow end-to-end.

Option 2: Use Your Own Data#

Same as option 1, except you collect your own teleoperation data and train your own models on the same vial-to-rack task.

Tip

This will take more time and work, but will help you experience the value of good demonstration data, watch how this affects policy performance, and more. Prove you can replicate the results independently.

Option 3: Bring Your Own Task#

Get more creative with this learning path as a base for your own exploration.

Buy the workspace, but swap out the props and task. Define a new manipulation problem, collect data for it, and apply the same sim-to-real strategies covered in this course.

Tip

This will take some creativity, or maybe you already have a task in mind. But it will ultimately teach you the most, to build and apply the process to a new task.

Option 4: Going Further#

Train a robust enough model that you can completely remove the lightbox enclosure and run the task in an uncontrolled environment.

Computer Hardware Prerequisites#

We have tested this workshop on:

Ubuntu Linux 24.04 with an RTX 5090 Laptop edition, 64GB RAM
Ubuntu Linux 24.04 with an RTX PRO 6000 Blackwell Workstation Edition, 125GB RAM

Details on the robot and workspace requirements can be found in Building the Workspace.

What’s Next?#

Continue to What is Sim-to-Real?.