Mobile Robot Controller Design by Evolutionary Multiobjective Optimization in Multiagent Environments

Evolutionary computation has been often used for the design of mobile robot controllers thanks to its flexibility and global search ability. A lot of studies have been done based on single-objective functions including weighted-sum scalarizing objective functions. For an example of mobile robot navigation, at least the minimization of the arrival time to the target and the minimization of dangerous situations should be considered. In this case, a weighted-sum of two objectives is always minimized. It is, however, difficult to specify an appropriate weight vector beforehand. This paper demonstrates the application of evolutionary multiobjective optimization to mobile robot navigation in order to optimize the conflicting objective simultaneously. We analyze the obtained non-dominated controllers through simulation experiments in multiagent environments. We also show the utilization of the obtained non-dominated controllers for situation change.