Application of semantic maps for mobile robot simulation

Abstract

In this paper a new concept of using semantic map for robot operator training purpose is described. The approach consists of 3D laser data acquisition, semantic elements extraction (using image processing techniques) and transformation to rigid body simulation engine, therefore the State Of the Art related to those research topics will be discussed. The combination of a 2D laser range finder with a mobile unit was described as the simulation of a 3D laser range finder in [1]. In this sense we can consider that several researches are using so called simulator of 3D laser range finder to obtain 3D cloud of points [2]. The common 3D laser simulator is built on the basis of a rotated 2D range finder. The rotation axis can be horizontal [3], vertical [4] or similarly to our approach (the rotational axis lies in the middle of the scanners field of view). Semantic information extracted from 3D laser data is recent research topic of modern mobile robotics. In [5] a semantic map for a mobile robot was described as a map that contains, in addition to spatial information about the environment, assignments of mapped features to entities of known classes. In [6] a model of an indoor scene is implemented as a semantic net. This approach is used in [7] where robot extracts semantic information from 3D models built from a laser scanner. In [8] the location of features is extracted by using a probabilistic technique (RANSAC). Also the region growing approach [9] extended from [10] by efficiently integrating k-nearest neighbor (KNN) search is able to process unorganized clouds of points. The semantic map building is related to SLAM (Simultaneous Localization And Mapping) problem [11]. Most of recent SLAM (Simultaneous Localization And Mapping) techniques use camera [12], laser measurement system [13] or even registered 3D laser data [14]. Concerning the registration of 3D scans described in [15] we can find several techniques solving this important issue. The authors of [16] briefly describe ICP (Iterative Closest Points) algorithm and in [17] the probabilistic matching technique is proposed. In [18] the mapping system that acquires 3D object models of man-made indoor environments such as kitchens is shown. The system segments and geometrically reconstructs cabinets with doors, tables, drawers, and shelves, objects that are important for robots retrieving and manipulating objects in these environments. A detailed description of computer based simulators for unmanned vehicles is shown in [19] [57]. Also in [20] the comparison of real-time physics simulation systems is given, where a qualitative evaluation of a number of free publicly available physics engines for simulation systems and game development is presented. Several frameworks are mentioned such as USARSim which is very popular in research society [21], Stage, Gazebo [22], Webots [23], MRDS (Microsoft Robotics Developer Studio) [24]. Some researchers found that there are many available simulators that offer attractive functionality, therefore they proposed a new simulator classification system specific to mobile robots and autonomous vehicles [25]. A classification system for robot simulators will allow researchers to identify existing simulators which may be useful in conducting a wide variety of robotics research from testing low level or autonomous control to human robot interaction.. To ensure the validity of robot models, NIST proposes standardized test methods that can be easily replicated in both computer simulation and physical form [26]. In this paper we propose a new idea of semantic map building, this map can be transformed into rigid body simulation. It can be used for several applications such as robot operator training. It is a new idea and can give an opportunity to develop training systems composed by real and virtual robots. We hope that it will improve multi robot system design and development. The paper is organized as follows: in section “Robot” robot and its data acq