The intelligent wheelchair mounted manipulator is a kind of assistive robots, which can providing both transporting and manipulating help. A program by demonstration paradigm for the guardians (nurses or family members) or the old and the disabled themselves is reasserted in this paper.An pattern recognition system is established to learn task execution from bad examples from the guardians or the old and the disabled. In the system, a modified mean squared velocity method is adopted to segment a trajectory into discrete trajectory subsection; then a nearest-neighbor classifier based on the transformative matching algorithm for curves recognition is adopted to classify the discrete trajectories with certain labels according to prior-approval templates, and so the source trajectory is turned into a string of labels; at last a string matching algorithm is adopted to find the effective part for task execution.To help the old and the disabled. accomplish a complex trajectory precisely, we develop a user adapted playing-back mode, which helping the user accomplish task execution by autonomically playing back demonstration trajectories. An bayesian networks is established to estimate user intent and plan, and then a demonstration trajectory (in object coordinate system) to accomplish user intent is played back autonomically.At last, the pattern recognition system and the teaching-playback mode is analyzed by simulation; an experimental platform of intelligent wheelchair is established, and an experiment of teach and playing-back is carried on to verify pattern recognition system and user adapted playing-back mode in this paper. |