Enhancing task classification in human-machine collaborative teleoperation systems by real-time evaluation of an agreement criterion

Human-machine collaborative teleoperation systems were introduced to overcome limitations of state-of-the-art teleoperation systems by using a virtual assistant that supports the human operator in the execution of a task. Since assistances are highly task-dependent a correct classification of the currently performed task is paramount. In this paper, we present a novel approach for improving task classification for a human-machine collaborative teleoperation system. Starting from a classical HMM-based classifier implemented in our previous research, we introduce a method for correcting erroneous task classifications by evaluating an agreement criterion. This criterion is based on interactive forces and is used to distinguish between situations in which human and assistant agree/disagree in their execution of the task. Using disagreement as indicator for the activation of an unsuitable/suboptimal assistance, erroneous task classifications are identified and the original classification result is revised. The proposed approach shows significant improvements in task classification coming along with a comparable low implementation effort.