 |
    |
Humanoid Robot Programming Based on CBR Augmented GP
Hongwei Liu1,2 and Hitoshi Iba1
1Graduate School of Frontier Science, The University of Tokyo, Hongo 7-3-1, Bunkyo-ku, Tokyo 113-8656, Japan
Lhw@iba.k.u-tokyo.ac.jp
Iba@iba.k.u-tokyo.ac.jp
2School of Computer and Information, Hefei University of Technology, Hefei 230009 China
Abstract. Humanoid robots are designed as companions for human beings to operate autonomously in various environments with people, and they need to adapt to noisy, cluttered environments. In order not only to look like but also to behave like human beings, humanoid robots need a vastly richer set of primitive behaviors; thus they must be able to produce dynamically temporal sequences of behaviors to accomplish a task in various human environments. As a consequence, it is di.cult to evolve control programs for a humanoid robot.
In this paper, as an improvement of our previous researches [3], we introduced adaptive and developmental mechanisms by augmenting Genetic Programming [2] with Case-Based Reasoning [1], thus endowing humanoid robots with online adaptive and developmental abilities. Experimental results show that this approach can generate robust control programs; although we use a highly simpli.ed simulation, which is rather crude, the robot can easily overcome the gaps between simulation and real world environments. Furthermore, the robot can develop new strategies according to the properties of new environments which it never encountered in simulation.
LNCS 3103, p. 708 f.
lncs@springer.de
© Springer-Verlag Berlin Heidelberg 2004