Imitation learning teaches an AI to act by copying examples of an expert doing the task, instead of learning only from rewards.
In plain words
The simplest form, behavioural cloning, treats expert demonstrations as a labelled dataset: for each situation, the right answer is what the expert did. It is easy but brittle. Once the learner makes a small mistake it lands in situations the expert never showed, and errors pile up. Methods such as DAgger fix this by asking the expert what to do in the situations the learner actually reaches.
A quick example
Imagine recording a person steering a toy car around a track, then training a network to map each camera frame to the steering they chose. If the car drifts somewhere new, DAgger-style training asks the person for the right move there too.
Where you'll meet it
- Training agents from recorded expert demonstrations.
- Improving behavioural cloning with extra expert feedback (DAgger).
Related
Sources
- DocsBehavioral Cloning (BC), imitation library docs (CHAI) · read 29 Sept 2026
- DocsDAgger, imitation library docs (CHAI) · read 29 Sept 2026
- PaperA Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning, Ross, Gordon and Bagnell (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.