Page - 49 - in Joint Austrian Computer Vision and Robotics Workshop 2020

Relay Final waypoint reached Stopping Replanning Unfeasible trajectory New global plan received No obstacle detected Replan requested Figure 2. Overview of the local planner state machine de- scribedasa statemachine 3.Architectureproposal In this section, we propose the architecture of our motion planner framework1. By investigating other planner frameworks we found the following typical characteristics: 1. The underlying planner is controlled by events (e.g. transition to re-planning, recovery initia- tion) in a fashion of a state-machine based ap- proach (a brief overview of the local planner statemachine is shownonFigure2). 2. Executionshouldnotstartbeforeall input infor- mation hasbeen received recently. 3. Transitions between states are not necessarily instantaneous, requiringsmootherswitchingbe- tween behaviors. The development of a (hybrid) timed state machine library was motivated by these properties, especially to efficiently resolve property 1 and 2. Hybrid au- tomata [1] is a well-studied way to model and ver- ify systems with both discrete and continuous timed properties and also to describe robot behaviors [2]. A behavior similar to what is presented on Figure 2 can be easily mapped to hybrid automata formalism. Transitions are either governed by discrete events and continuous activities. For example, a continu- ous variable in this case could be the distance to the closest obstacle detected and a typical discrete event is the request to replan a segment of the trajectory or to execute fallback scenario. We ensure, that each transitionsarepublishedtoamiddlewareframework, enablingversatile runtimeverification. Our goal was to follow a highly-distributed archi- tecture, where sub-tasks are decomposed from the planner component. In our approach, the perception relatedtaskslikeobstacledetectionandclassification are decomposed from other specific planner tasks. 1Availableathttps://github.com/kyberszittya/hotaru planner.git This enables the reuse of components and isolated verification. Perception components are interacting withplannercomponentsbyinducingdiscreteevents and modifying continuous signals. For instance, an obstacle detection component may trigger the local planner to replan by raising a discrete-timed event. After theobstacle isavoided, theplanner restores the remainderof theoriginal trajectory in relaymode. 4.Conclusion In conclusion, we provided an overview of a new motion planning framework under development which can be easily extended with new algorithms and tuned to specific domain requirements. A new initial motion planner framework version is created. The extension of our framework with various local planner methods is a primary focus. Global trajec- tory planner methods will be integrated in the fu- ture. Our automata framework and the related code- generator toolwill bealsoenhanced. 5.Acknowledgments This work was supported by the Hungarian Gov- ernment and the European Union within the frames of the Sze´chenyi 2020 Programme through grant GINOP-2.3.4-15-2016-00003 References [1] R. Alur, C. Courcoubetis, N. Halbwachs, T. A. Hen- zinger, P.-H. Ho, X. Nicollin, A. Olivero, J. Sifakis, and S. Yovine. The algorithmic analysis of hybrid systems. Theoretical Computer Science, 138(1):3 – 34,1995. [2] M. Egerstedt, K. Johansson, J. Lygeros, and S. Sas- try. Behavior based robotics using regularized hybrid automata. In Proceedings of the IEEE Conference on Decision and Control, volume 4, pages 3400 – 3405 vol.4, 1999. [3] S. Kato, S. Tokunaga, Y. Maruyama, S. Maeda, M. Hirabayashi, Y. Kitsukawa, A. Monrroy, T. Ando, Y. Fujii, and T. Azumi. Autoware on Board: En- abling Autonomous Vehicles with Embedded Sys- tems. In 2018 ACM/IEEE 9th International Confer- enceonCyber-PhysicalSystems(ICCPS),pages287– 296,Apr.2018. [4] S.M.LaValle. PlanningAlgorithms. CambridgeUni- versity Press,NewYork, NY,USA,2006. [5] S. Pu¨tz, J. S. Simo´n, and J. Hertzberg. Move Base Flex: A Highly Flexible Navigation Framework for MobileRobots. In2018IEEE/RSJInternationalCon- ference on Intelligent Robots and Systems (IROS), Oct. 2018. 49