Page - 48 - in Joint Austrian Computer Vision and Robotics Workshop 2020

ReactiveMotionPlanningFrameworkInspiredbyHybridAutomata CsabaHajdu, A´ronBallagi Sze´chenyi Istva´nUniversity {hajdu.csaba,ballagi}@sze.hu Abstract. This paper presents a motion planning framework controlled by reactive events and pro- ducing feedback data suitable to be processed by various learning and verification methods (e.g. re- inforcement learning, runtime monitoring). Our architecture decomposes subtasks of motion plan- ning into separate perception and trajectory plan- ner parts. In our architecture, we interact between these distributed parts through discrete-timed events controlled by timed state machines, besides classi- cal continuousstateflow. Ourresearchprimarily fo- cusesonautonomousvehicleresearch,so this frame- work is supposed to satisfy the requirements of this field. The motion planner framework interfaces a widely-used roboticmiddleware. 1. Introduction Motion planning (or trajectory planning) is a mandatory task both in mobile robotics and in au- tonomous vehicle navigation [4]. The field has been actively researched and used, providing efficient al- gorithmssuitable fordifferentdomainsandrobot se- tups. The role of motion planning in robotics is to create a feasible, collision-free path between the lo- cation of the agent (mobile robot or vehicle) and an arbitrarily defined goal point, based on the agent’s sensory input and actuation. On the other hand, the emergenceofautonomousvehiclesandother special UGVs requires high-reliability, computational effi- ciency and optimization of velocity profile even in rough environmental conditions. The typical problems of motion planner frame- works are their relatively hard extension and limited verification capabilities. In this paper, we propose a prototype of a motion planning architecture with the focus on providing comprehensive verification out- put and extensioncapabilities. Figure1.Electronicallymodifiedautonomous testvehicle 2.Motivationandrelatedwork Thedevelopmentofanewmotionplanningframe- work was motivated by ongoing research at our uni- versity. We are developing an autonomous vehi- cle (anelectronicallymodifiedNissanLeafequipped with numerous sensors, Figure 1) and a differential drive robot in various projects. Both rely on motion planning, thus our aim is to create a motion planner frameworkusable inbothapplication -withminimal configurationeffort. Many commercially available unmanned ground vehicles (UGV) use ROS and its integrated naviga- tion component, move base. This framework is a monolithic implementation with plugin-oriented ex- tension and occupancy grids as a basis of environ- ment representation. Some of these issues had been addressed in move base flex [5]. In autonomous ve- hicle frameworks, Autoware [3] provides a loose ar- chitecture enabling the replacement of its built-in motionplanningcomponentwithdifferent solutions. In both approaches, reactive events (e.g. synchro- nization of all incoming topics, the transition to re- planning, etc.) in both systems are relatively hard to traceanddebug. 48