Effective diversity maintenance in deceptive domains

摘要: Diversity maintenance techniques in evolutionary computation are designed to mitigate the problem of deceptive local optima by encouraging exploration. However, as problems become more difficult, heuristic fitness may increasingly uninformative. Thus, simply genotypic diversity fail much increase likelihood evolving a solution. In such cases, needs be directed towards potentially useful structures. A representative example search process is novelty search, which builds rewarding behavioral novelty. this paper effectiveness fitness, novelty, and objectives compared two robotics domains. biped locomotion domain, helps evolve control policies that travel farther before falling. best method optimize objective together. maze navigation ineffective while still increases performance. The conclusion works well-posed domains, phenotypic information, like necessary for highly ones.