Autonomous Robotic Assembly with Variable Material Properties

Abstract

This paper discusses the problems within autonomous robotic assembly workflows as they encounter a variable property of assembly parts or materials. This is shown through a case study with an industrial robot in an enclosed work cell and a simple assembly task with wooden sticks of variable lengths, designed as an adaptive feedback control system. To perform the study, the development of a virtual model for the persistent storage of material data and computation of next build-actions is required. Different sensing strategies are used to address issues of substantial, and minute, material variabilities of dimensional properties as they deviate from a predictive virtual model. Establishing communication strategies for a live-control pipeline as the infrastructure for this system allows the system to respond to pre-build scans of part dimensions, as well as update the virtual model when post-build scanning detected deviation. In the worst-case scenario—if preconditions were not met—the post-build scan would be unsuccessful and the system would self-terminate. Otherwise, deviations would update and influence future actions. This influence is what leads to the indeterminate nature of the resultant forms.