Observability Properties of Object Pose Estimation

This is the code for the paper

Observability Properties of Object Pose Estimation

Trevor Avant & Kristi A. Morgansen, American Control Conference (ACC) 2019

code

Before running this code you should install Blender, add this repository to your PYTHONPATH (because the pose_estimation directory is structured as a Python package), and change the directories in pose_estimation/directories.py to directories on your computer.

• Figure 1: run python pose_estimation/gramian/example.py

  • this figure was generated by letting name = cone, and changing render_props.eps to several different values

• Figure 2: run python pose_estimation/gramian/epsilon.py

• Figure 3: run python pose_estimation/gramian/example.py

  • uncomment name = ... for different models

• Figure 4: run python pose_estimation/gramian/best_views.py

• Figure 5: run python pose_estimation/gramian/dynamic_object.py

• Figure 6: run python pose_estimation/gramian/trajectories.py and then blender --background --python pose_estimation/gramian/trajectories_plot.py

  • you will have to run the second command twice: once with draw_opt_curves('min') uncommented at the end, and once with draw_opt_curves('max') uncommented

Blender models

The Blender models for the chair, lamp and car are originally from blendswap.com (and we made some modifications to the original models). All of the original models are licensed under Creative Commons CC BY licenses which permit sharing and adaptation if attribution is given. The links to the original models are:

• chair
• lamp
• car

Note that if other .blend models are to be used in this project, the 3D model must be part of a scene named "Scene". Additionally, if the model in consists of multiple objects, all objects should be parented to an object called "all_parts".