Jiming Liu of Hong Kong Baptist University and Shiwu Zhang of the University of Science and Technology of China have published a new paper on Multiphase Genetic Programming in the International Journal of Pattern Recognition and AI. The authors describe a genetic algorithm approach to evolving maneuvers for a sumo robot called Junior. Unlike prior papers on the subject, this one used a real robot rather than a simulation. IJPRAI is a proprietary journal that doesn't allow free access to articles but you can see an postscript version of the full paper on the HKBU website along with related information about multi-agent simulation of complex adaptive systems. The same authors also have a prior paper (PS format) on Multi-Phase learning in Sumo robots.