This work addresses the fundamental problem of how a robot acquires local knowledge about its environment. The domain that we are concerned with is a speech-commandable robotic wheelchair operating in a home/special care environment, capable of navigating autonomously to a verbally-specified location in the environment. We address this problem by incorporating a narrated guided tour following capability into the autonomous wheelchair. In our method, a human gives a narrated guided tour through the environment, while the wheelchair follows. The guide carries out a continuous dialogue with the wheelchair, describing the names of the salient locations in and around his/her immediate vicinity. The wheelchair constructs a metrical map of the environment, and based on the spatial structure and the locations of the described places, segments the map into a topological representation with corresponding tagged locations. This representation of the environment allows the wheelchair to interpret and implement high-level navigation commands issued by the user. To achieve this capability, our system consists of an autonomous wheelchair, a person- following module allowing the wheelchair to track and follow the tour guide as s/he conducts the tour, a simultaneous localization and mapping module to construct the metric gridmap, a spoken dialogue manager to acquire semantic information about the environment, a map segmentation module to bind the metrical and topological representations and to relate tagged locations to relevant nodes, and a navigation module to utilize these representations to provide speech-commandable autonomous navigation.
展开▼
