Online Intention Recognition in Computer-Assisted Teleoperation Systems

Limitations of state-of-the-art teleoperation systems can be compensated by using shared-control teleoperation architectures that provide haptic assistance to the human operator. This paper presents a new approach for computer-assisted teleoperation, which recognizes human intentions and dependent on the classified task activates different types of assistances. For this purpose, time series haptic data is recorded during interaction, passed through an event-based feature extraction, and finally used for task classification by applying a Hidden Markov Model approach. The effect of the assistance function on human behavior is discussed and taken into account by training multiple classifiers for each type of assistance. The introduced approach is finally validated in a real hardware experiment. Results show an accurate intention recognition for assisted and non-assisted teleoperation.