Logistics

• People are taking too much advantage of the free extra three days, and then often even asking for an extension on that.
• I feel very generous, but that is causing havoc in our grading.
• So:

Standups

• A weekly new deliverable beteween now and the end of the term: Weekly Standups
• The first one will be due next Tuesday

Lab Time

1. To the student lab managers, please make sure you have 2 hours per week on the calendar
2. If you have a problem with a robot that you cannot solve, please submit a Lab Service Request

Introduction to Fiducials

• A specific kind of graphical image combined with an algorithm
• Generally the recognition algorithm includes a tool to create the images

• Standardized image with known dimensions
• Conceptually simple geometrical transformation
• Perspective distortion is used to compute relative pose (orientation and distance)
• Tricky business with TFs and POV!

Aruco Fiducials

• The basic concept of Fiducials can be implemented in different ways
• We use aruco fiducials. The other well known one are apriltags
• They each have slightly different performance characteristics
• aruco fiducials are supported directly by OpenCV
• We have also used Ubiqutiy Robotics aruco_detect

The Software

• More generic CV: OpenCV aruco detection
• ROS Oriented: aruco_detect
• They are both useful depending on what you are doing

Characteristics of Aruco Fiducials

• Easily recognized by camera using computer vision
• Known dimensions
• Known pose (location in 6d) within a certain coordinate system
• Encode a number so that your software can tell them apart if there is more than one

Creating Aruco Fiducials

• Lots of ways, in code or online
• e.g. Aruco Markers Generator
• You can choose the size, the number of bits, and the number of distinct tags

Tag Detection

• How do we get the transform between a camera and fiducials?
• Compare the orientation, size and transfor betwen the obsrved image and the known features of the fiducial
• ROS Wiki with demo

Tag Placement

• Trade-offs: along the wall, along the ceiling
• How many tags do you need?
• Can you have too many?
• Remember: the tag is not identifying a thing
• It is just a point where your real world coordinate system is bound to your robot coordinate system

Fiducial Localization

• In general localization means computing the transform between real world and robot coordinates.
• More specifically we are computing a transform betwen a map tf and a robot_base tf
• Even more specifically what is the transform between the pose of the fiducial and the pose of the camera.

Lab Notebook Entries

• A coming homework asks that you write a “howto” for our lab notebook
• Here’s one about Localization with Fiducials from another semester

Other Resources

• fiducial_follow

• Ubiquity Robotics Fiducials

• Two types of Fiducial Localization

Fiducial SLAM

• Detect fideucials in image
• Compute transform between fiducial(s) and camera (and robot)
• Orient a coordinate system corresponding to the real world

Fiducial Localization with an existing map

• Predefine a map with fiducials marked on it
• Detect fiducials in camera image
• Find transforms between fiducials and camera
• Localize robot on that map

Thank you. Questions?  (random Image from picsum.photos)