In this paper, we study the problem of using a planning algorithm to automatically create and update a Behavior Tree (BT), controlling a robot in a dynamic environment. Exploiting the characteristic of BTs, in terms of modularity and reactivity, the robot continually acts and plans to achieve a given goal using a set of abstract actions and conditions. The construction of the BT is based on an extension of the Hybrid Backward-Forward algorithm (HBF) that allows us to refine the acting process by mapping the descriptive models onto operational models of actions, thus integrating the ability of planning in infinite state space of HBF with the continuous modular reactive action execution of BTs. We believe that this might be a first step to address the recently raised open challenge in automated planning: the need of a hierarchical structure and a continuous online planning and acting framework. We prove the convergence of the proposed approach as well as the absence of deadlocks and livelocks, and we illustrate our approach in two different robotics scenarios.
from cs.AI updates on arXiv.org http://ift.tt/2eZcqEZ
via IFTTT
No comments:
Post a Comment