Use context to understand user's implicit intentions in Activities of Daily Living

For an assistive robot, accurately understanding the user's intention with minimal user involvement is critical in human-robot interaction. Especially for the elderly in their Activities of Daily Living (ADL), implicit intention recognition using subtle cues could provide them great convenience as well as improving quality of life. Usually user intentions in ADL are closely connected with specific context. For example, the user's intention for a cup on a hot day is likely to be drinking. Therefore, the relation between intention and context should be emphasized. The goal of our research is to effectively infer implicit human intention using the object that the user attends to and its subtle context. The context helps to filter the coarse intentions. Towards this goal, a new context-specific implicit intention recognition (IR) model has been established based on a Bayesian Network (BN) algorithm. Using set of questionnaires, the strongly intention-related context features have been selected and their conditional values for different intentions have been calculated. Then another set of questionnaires were used to verify the model's effectiveness. The results showed the BN based context-specific implicit IR model could help the robot to proactively get a good understanding of a user's implicit intentions with merely subtle context cues.