Assembly Status Detection Based on Evaluation of not Mounted Parts

In a human robot collaboration assembly scenario the human might not follow a given assembly plan for various reason. Therefore the robot is required to adapt it's behavior to those unpredictable situations. Normally this would require a lot of manual input by a human to reprogram the robot according to this new situation. The focus of this paper is a fully autonomous detection of the current assembly state of an object by just observing the not yet mounted parts. Based on the information of the not mounted parts present in the scene and all possible assembly sequences for a given object, stored in an AND/OR Graph, a traversal algorithm is proposed to determine all possible assembly states and decide on the most likely one. Furthermore, we propose an extension of the initial algorithm to handle non distinct parts. Last but not least the solutions is evaluated in 4 different assembly scenarios.