Human-Adaptive Robot Interaction Using Interactive EC with Human-Machine Hybrid Evaluation

摘要

Using interactive evolutionary computation (IEC), we created human-robot interaction system that maintains user interest over time. Although IEC enables users to design systems reflecting subjective preferences, it forces them to evaluate a large number of individuals. The refined IEC techniques, we propose in this regard, human-machine hybrid evaluation (HMHE), lets users manually evaluate only representative genes, after which HMHE automatically estimates the fitness of other genes, thereby increasing a population without increasing user evaluation process. Experimental results showed that preferences easily change in interaction. We confirmed that HMHE maintains high diversity, while maintaining user interest.