Robots typically operate either (a) based on static programs, (b) on optical recognition of parts. (b) can is typically solved through expensive hardware-based solutions that require deterministic lighting and contrast conditions, and configuration. These solutions are typically employed for high-volume production environments. An other solution vector for (b) is to employ machinelearning solutions. Training such solutions is hard, and computing intensive, especially for scenarions with many variantions. While solutions might either detect individual parts on an AGV, a different approach is to detect the position of the AGV. For low-volume flexible production scenrios it is imperative that new parts, new variations and new scenarios can be integrated/automation with as low effort as possible, by normal factory workers instead of IT specialists.
The purpose of this bachelor thesis is to develop a prototype to detect the position of AGVs based on positional markers (see https://en.wikipedia.org/wiki/Fiducial_marker), with a special focus on integration into a process-based environment that allows for easy modelling different detection scenarion and integration with static robot programming. The thesis includes the creation of a software prototype. *vi* users preferred.
Contact: bachelor.i17 [at] in.tum.de