Proceedings of the 4th International Conference on Autonomous Robots and Agents, Feb 10-12, 2009, Wellington, New Zealand Local WMR Navigation with Monocular Data Lluis Pacheco, Ningsu Luo, Xavier Cuft and Javier Cobos Depts of Computer Engineering and Electronics and Control Engineering University of Girona Girona, Spain {lluispa,ningsu,xcuf,jcobosg}@eia.udg.edu Abstract- This article presents recent WMR (wheeled mobile robot) navigation experiences using local perception knowledge provided by monocular and odometer systems. A local narrow perception horizon is used to plan safety trajectories towards the objective. Therefore, monocular data are proposed as a way to obtain real time local information by building two dimensional occupancy grids through a time integration of the frames. The path planning is accomplished by using attraction potential fields, while the trajectory tracking is performed by using model predictive control techniques. The results are faced to indoor situations by using the lab available platform consisting in a differential driven mobile robot. Keywords-Mobile robot navigation, local path planning, model predictive control, monocular perception techniques I. INTRODUCTION Autonomous robot navigation without GPS is an interesting research topic faced by navigators [1]. Therefore, indoor WMR (wheeled mobile robot) navigation using computer vision techniques are normally based on environment description. The data analysis can be considered as a map when it is provided enough information to reach robot localizations. Normally, navigation based on maps includes geometric environmental information. The first research was developed using 2D environment models consisting of a free or occupied grid [2]. The virtual potential field idea improved performance by using objective attraction and obstacle repulsion forces [3]. Therefore, WMR navigation strategies should be planned so that obstacle collisions are avoided during navigation towards the objective. The meaningful navigation idea consists of providing feasible and expected visual landmarks, thereby allowing the robot's map position to be calculated using information about recognized landmarks. Localization of the WMR using machine vision system computation is done by using sensor information from where landmark detection is correlated with the corresponding map position. Then, the robot position and orientation are estimated using the previous data. WMR localization can be considered as absolute or incremental [4]. Incremental navigation methods consider the initial robot coordinates as known; machine vision system information is used as a way to improve the robot's positioning. Otherwise, absolute navigation methods would not know the initial robot coordinates. When absolute localization is used, the navigation system should build a correspondence between the WMR's perception information and the database. Estimations based on sensor uncertainty probability allow better robot localization. The ambiguities can be solved by using statistical methods [5]. The incremental localization can use three different techniques: localization based on geometrical representation, localization based on topological space representation, or localization based on landmark detection [6],[7] and [8]. The different presented methods differ in the map representation however all share the need of landmark search and detection that allow computing the robot localization. Furthermore, other research work studying natural agents has presented a new robot navigation paradigm. Hence, corridor planning for natural agents has been presented as a new and useful robot control and planning framework using low level obstacle avoidance and simple control [9]. Butler has developed a system in which innate local navigation animal abilities are combined with a very simple imposed path-like structure to produce a desired overall motion. The developed research showed that local animal abilities such as control and perception can be combined with a very simple imposed path-like structure to produce the desired overall motion. Hence, path-planning in partially unknown environments should be short enough to allow local reactive behaviours; consequently long trajectory planning should be flexible in order to deal with short term uncertainties while the global task is accomplished. This article explores this challenge as a navigation strategy for the available WMR platform PRIM that consists in a differential driven one with a free rotating wheel [10]. Hence, in this work, local animal abilities are performed by using a simple perception system, which consists of a monocular camera and an odometer system. Moreover, the field of view is constrained to the WMR neighbourhood; thus only few seconds of trajectory planning can be done. The simple corridor structure is faced up by using local attraction potential fields; where MPC (model predictive control) techniques are used to perform accurate trajectory tracking. Therefore, an example in indoor scenarios and the path followed is reported. 978-1-4244-2713-0/09/$25.00 ©2009 IEEE 584 Authorized licensed use limited to: UNIVERSITAT DE GIRONA. Downloaded on May 8, 2009 at 06:53 from IEEE Xplore. Restrictions apply.