Towards workflow acquisition of assembly skills using Hidden Markov Models

In recent years, the demand for efficient systems which can capture and learn human skills has become increasingly important. In this paper an approach for acquiring and recognizing human assembly skills is presented. The underlying workflows of assembly skills are captured by using a simple multi-sensor data glove and camera tracking. To avoid the processing of redundant information, at first the relevant tasks of a workflow are identified by analyzing measuring information of the multi-sensor capturing system. Thus, only relevant data is comprised in the representation of a workflow. Unlike common approaches a workflow is modeled as entire unit using a continuous hidden Markov model (HMM). The recognition process of input patterns is based on an adaptive threshold model that can identify known workflow patterns and non-meaningful input patterns as well.