Applying Real-Time Survivability Considerations in Evolutionary Behavior Learning by a Mobile Robot

In this chapter we investigate real-time extensions for evolutionary mobile robot learning. The learning performance is measured in navigation experiments of complex environments as performed in a Kephera mobile robot simulator (YAKS). All these experiments are done in the context of our recently introduced motivation based interface that provides an intuitive human-robot communications mechanism (Arredondo et al., 2006). This motivation interface has been used in a variety of behavior based navigation and environment recognition tasks (Freund et al., 2006). Our first heuristic introduces active battery level sensors and recharge zones, which are used as soft deadlines to improve robot behavior for reaching survivability in environment exploration. Based on our previously defined model, we also propose a hard deadline based hybrid controller for a mobile robot, combining behavior-based and mission-oriented control mechanisms. These methods are implemented and applied in action sequence based environment exploration tasks in a YAKS mobile robot simulator. We validate our techniques with several sets of configuration parameters on different scenarios. We consider soft-deadlines as a dangerous but not critical battery charge level which affect a robot's fitness. Harddeadlines are considered as a possible (because of partial knowledge) point where, if the robot does not recharge his battery, an unrecoverable final freezing state is possible. Our tests include action sequence based environment exploration tasks. These experiments show a significant improvement in robot responsiveness regarding survivability and environment exploration when using these real-time based methods. The rest of the chapter is organized as follows. In Section 2 a description of our softcomputing based navigation model is given. In Section 3 real-time extensions of our model are presented. In Section 4 we show the experimental setup and test results. Finally, in Section 5 some conclusions and future work are drawn.