Skip to content

Add sensitivity analysis example workflow#866

Draft
cvolkcvolk wants to merge 2 commits into
mainfrom
cvolk/docs/sensitivity-analysis-workflow
Draft

Add sensitivity analysis example workflow#866
cvolkcvolk wants to merge 2 commits into
mainfrom
cvolk/docs/sensitivity-analysis-workflow

Conversation

@cvolkcvolk

Copy link
Copy Markdown
Collaborator

Summary

Document variation and sensitivity workflows

Detailed description

  • Add a two-part example covering environment variations and camera sensitivity analysis.
  • Include runnable DROID configurations, visual output, report interpretation, and practical insights.
  • Add the workflow to the documentation navigation and correct the NPE description.

Signed-off-by: Clemens Volk <cvolk@nvidia.com>

@alexmillane alexmillane left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks great!

One small suggestion!

Comment on lines +123 to +147
.. todo::

Replace this dry-run figure with the final plot from an evaluation that contains enough
successful and failed episodes to support an interpretation.

.. figure:: ../../../images/droid_camera_sensitivity_dry_run.png
:width: 100%
:alt: Dry-run sensitivity report for three wrist-camera translation components
:align: center

Temporary report from the five-episode dry run. Horizontal axes show offsets in metres. The
blue curve is the estimated distribution for successful episodes, its shading marks the 5%
to 95% range, and the grey dashed line is the uniform sampling distribution. A blue curve
close to the dashed line suggests no clear relationship; concentration in one region
suggests a stronger association with success.

As an illustration of how to read the shapes, imagine that the same pattern came from a larger
evaluation containing both successes and failures. The takeaways could then be written as:

* **Horizontal displacement:** Pi0.5 appears more successful when the wrist camera moves to the
right of its nominal position.
* **Vertical displacement:** Pi0.5 appears relatively robust to vertical movement because the
posterior remains close to the uniform reference.
* **Depth displacement:** Pi0.5 appears more successful when the wrist camera moves about 3 mm
forward from its nominal position.

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

What do you think about including in the repo a jsonl file from a larger evaluation.

That way we could show some meaningful results and a user could run the same analysis?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants