This paper presents a simple approach for the acquisition and representation of spatial knowledge needed for controlling a
semi-autonomous wheelchair. Simplicity is required in the domain of rehabilitation robotics because typical users of assistive
technology are persons with severe impairments who are not technical experts. The approach is a combination of carrying out
so-called
basic behaviors and the analysis of the wheelchair’s
track of motion when performing these behaviors. As a result, autonomous navigation in the user’s apartment or place of work can be learned
by the wheelchair by teaching single routes between potential target locations.
This paper focuses on the analysis of the motion tracks recorded by the vehicle’s dead reckoning system. As a means for unveiling
the structure of the environment while the system is moving, an incremental generalization is applied to the motion tracks. In addition, it is discussed how two of such generalized motion tracks are match to perform
a one-dimensional self-localization along the route that is followed.