Distributed Architecture for Intelligent Robotic Assembly Part II: Design of the Task Planner

Abstract

mercial devices have emerged in recent years to aid industrial applications Active compliance can be roughly divided into fine motion planning and reactive control. Fine motion planning relies on geometrical path planning whereas reactive control on the synthesis of an accommodation matrix or mapping that transform the corresponding contact states to corrective motions. A detailed analysis of active compliance can be found in Perhaps, one of the most significant works in fine motion planning is the work developed by Lozano-Perez, Mason and Taylor known as the LMT approach The LMT approach automatically synthesizes compliant motion strategies from geometric descriptions of assembly operations and explicit estimates of the errors in sensing and control. Approaches within fine motion planning can also be further divided into model-based approaches and connectionist-based approaches though, some reactive control strategies can be well accommodated within the model-based approach. In either case, a distinctive characteristic in model-based approaches is that these take as much information of the system and environment as possible. This information includes localisation of the parts, part geometry, material types, friction, errors in sensing, planning, and control, etc. On the other hand, the robustness of the connectionist-based approaches relies on the information given during the training stage that implicitly considers all the above parameters. In this chapter we present a "Task Planner", connectionist-based approach that uses vision and force sensing for robotic assembly when assembly components geometry, location and orientation is unknown at all times. The assembly operation resembles the same operation as carried out by a blindfold human operator. The task planner is divided in four stages as suggested in