Improving the continuous execution of reactive LTL-based controllers

Abstract

Recently, formal methods have been used to transform high-level robot tasks into correct-by-construction controllers. While correctness is guaranteed, these inherently discrete methods often lead to behaviors that are not optimal in the continuous sense, i.e. they induce robot paths that are significantly suboptimal. This paper proposes an algorithm for dynamically reordering the robot goals and connecting them via the shortest path with respect to a given continuous metric. The generated robot trajectories are close-to-optimal while satisfying the task specification in a dynamic environment. This method is implemented and simulation results are shown.