robomimic core team · Technical tutorial

Reproducing Published Experiments and Results

A reproduction guide that connects dataset preparation with the training, logging and experiment instructions needed for comparison.

Source text reviewed October 27, 2026. Study notes and exercises are written by botstest.

Read the original ↗

Where to start

Complete the linked training and results tutorials before attempting a published reproduction.

Use it in your project

Record the dataset pipeline, configuration and evaluation choices for a small comparison.

Scope and limitations

A guide to a particular framework; reproduce the referenced protocol before claiming comparable results.

Follow a related roadmap

Run a reproducible manipulation experiment · Intermediate →

Read: Complete the linked training and results tutorials before attempting a published reproduction.

Use the reproduction prerequisites to distinguish a comparable experiment from a loosely similar demonstration.

Produce: Keep the dataset pipeline, training configuration and evaluation conditions fixed for a comparison.

Collect demonstrations and train an imitation policy · Professional →

Read: Complete the linked training and results tutorials before attempting a published reproduction.

Use the reproduction prerequisites to distinguish a comparable experiment from a loosely similar demonstration.

Produce: Release the exact dataset processing, configuration and evaluation instructions used for each checkpoint.

Evaluate vision-language-action models responsibly · Professional →

Read: Complete the linked training and results tutorials before attempting a published reproduction.

Use the reproduction prerequisites to distinguish a comparable experiment from a loosely similar demonstration.

Produce: Trace every reported result to model, data, environment and evaluation configuration versions.