In this paper a preliminary -work towards grounded concept learning for a service robot through its vision and human interaction is presented. With a lifelong learning server (ILL), described in [8], the robot can incrementally learn to recognize instances of such concepts of indoor objects as "Person", "Trash-can" and "Triangle sign" using simple intra-band statistical features extracted from the Haar wavelet transform of its vision information under the instruction of a human teacher. Experimental results show that these simple wavelet-based features can efficiently describe the characteristics of different objects in an office-like environment. Comparison with some other feature extraction methods is also given.
展开▼