Scalable Low-Cost Photogrammetry for Enhanced In-the-Loop Robotic Deployment – Experimental Snapshot and Future Potential

Abstract

On-line metrology systems capable of rapid 6DoF computation are proven enablers for increasing

accuracy and metric performance of robotic control systems. Applications range from improved tool

path delivery to the accurate placement and alignment of sensors used for inspection of as-

manufactured components. However, present solutions are hard to adapt to new robotic cell

configurations easily, making them difficult to use across the range of application use cases. This

presentation explores novel options for the flexible deployment of low-cost off-the-shelf camera

systems in photogrammetric networks capable of on-line deployment across small, medium, and

large volumes matching state-of-the-art industrial tools.

Starting at small scales with a KR6 robot system with a working volume of 1.36m3, the development

of a highly adaptive multi-camera system is detailed through experimental data gained from over 1

million cycles of 6DoF robot tracking that start to represent the challenges presented by a wide

range of manufacturing and inspection tasks. Emphasis is on avoiding large scale stable structures

and tooling, keeping systems flexible and rapid to deploy, calibrate and operate. Validation against

simultaneously acquired laser tracker end effector localisation connects estimated 6DoF solutions

with achievable accuracy.

Fig 1. Left: KR6 experimental setup with path planned “donuts” Right: low-cost photogrammetrically

tracked “real donuts” running into industry standard SA software.

This presentation extends to utilisation of the same systems for sub-mm robotic tracking at larger

volumes demonstrated with a highly unique KR500 mounted OCRobotics snake robot inspection

system with an 8m reach and a multitude of degrees of freedom. Here small-scale solutions

achieved with the KR6 bench level tests are expanded into low-cost imaging solutions that can be

deployed for inspection of large scale manufactured assemblies where networks of close-range

robot mounted sensors can be brought into and tracked in close proximity for wide area surface

inspection tasks. In doing so, this work demonstrates the flexibility of the presented solution to

maintain state-of-the-art accuracy and metric performance deployed against highly different robotic

scenarios.

Fig 2. Tracking camera view of “sleeping” 8m reach snake inspection robot. One of a network of 9

convergent views acquired from roof mounted low-cost photogrammetric cameras at a “safe” 12m

range.