Abstract:

Human-robot interaction (HRI) is an important topic in robotics, especially in assistive robotics. In this paper, we addressed the HRI problem in a smart assisted living (SAIL) system for elderly people, patients, and the disabled. Two problems were sloved that are very important for developing natural HRI: hand gesture recognition and daily activity recognition. For the problem of hand gesture recognition, an inertial sensor is worn on a finger of the human subject to collect hand motion data. A neural network is used for gesture spotting and a two-layer hierarchical hidden Markov model (HHMM) is applied to integrate the context information in the gesture recognition. For the problem of daily activity recognition, two inertial sensors are attached to one foot and the waist of the subject. A multi-sensor fusion scheme was developed for recognition. First, data from these two sensors are fused for coarse-grained classification. Second, the fine-grained classification module based on heuristic discrimination or hidden Markov models (HMMs) are applied to further distinguish the activities. Experiments were conducted using a prototype wearable sensor system and the obtained results proved the effectiveness and accuracy of our algorithms.

Key words: human-robot interaction, hidden Markov model, neural networks