ABSTRACT
In this paper we address the problem of defining survey trajectories for robotic platforms equipped of short range perception sensors and using dead-reckoning navigation, explicitly considering the impact of positioning errors in the execution of mapping missions. Assuming that the mapped region is delimited by a set of known objects, each possible exploratory trajectory is evaluated not only with respect to path length and surface coverage but also with respect to the effective utilization of the border of the explored region. We apply stochastic optimization tools (genetic algorithms) to find a set of trajectories which correspond to distinct trade-offs between the individual criteria. New appropriate reproduction and mutation algorithms have been defined, which must deal with the fact that the solution space is the union of several finite dimensional spaces. To cope with the multi-criteria nature of the problem, we use an evaluation/selection strategy based on the notion of Pareto set, which tries to identify the set of non-dominated solutions of the problem. We present results for ideal and real situations, which show that the genetic algorithms are able to find solutions with geometric regularities, learning basic local “filling modes” dependent on the geometry of the region.
