M30.5 CONNECT THE MECHANISM
Learn physical actions from demonstrations
Half your demonstrations steer left of the fruit bowl and half steer right. Train a robot to copy them the obvious way and it drives straight into the bowl. Here's why, and the fixes.
LESSON OVERVIEW14 min lesson
Lesson overview
Half your demonstrations steer left of the fruit bowl and half steer right. Train a robot to copy them the obvious way and it drives straight into the bowl. Here's why, and the fixes.
What you’ll explore
- Explain how robot demonstrations are recorded, why behavior-cloning errors compound and how action chunking and expert corrections help, and why averaging multimodal demonstrations fails while generative policies such as Diffusion Policy succeed.
GO TO THE SOURCE
Original explanations, connected to the research.
Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (ALOHA and ACT; Zhao et al., 2023)Diffusion Policy: Visuomotor Policy Learning via Action Diffusion (Chi et al., 2023)A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning (DAgger; Ross, Gordon & Bagnell, 2011)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.