Autonomous Exploration of Large Unknown Indoor Environments for Dense 3D Model Building Ivan Maurović ∗ Marija Ðakulović ∗ Ivan Petrović ∗ ∗ University of Zagreb, Croatia, Faculty of Electrical Engineering and Computing, Department of Control and Computer Engineering (email: ivan.maurovic@fer.hr, marija.dakulovic@fer.hr, ivan.petrovic@fer.hr) Abstract: Autonomous exploration and mapping of indoor environments is important task for building inspections. Mapping of large environments in 3D requires high memory and computational consumptions. In this paper, we present a 3D exploration strategy for a mobile robot equipped with a 3D laser scanner. Our strategy does not require a map of the environment and ensures on-line exploration of large unknown spaces. We propose a room detection algorithm and focus on the room-by-room exploration keeping the memory and computational requirements low. We evaluated our strategy by simulations and experimentally using a real mobile robot. Keywords: Autonomous exploration, 3D model, 2D exploration, 3D exploration, Room detection 1. INTRODUCTION In this paper we present a 3D exploration strategy for large indoor environments. A mobile robot equipped with a 3D laser scanner autonomously navigates through the environment with the aim to build its dense 3D model. The exploration strategy refers to obtaining exact, discrete scanning positions from which an unknown environment is completely modeled. The number of positions should be as minimal as possible to decrease the exploration time. The process of environment exploration finds the applications in the area of modeling indoor environments for different purposes such as building inspections, structural inspec- tions or in a combination with a thermal camera, the model can provide complete thermal information and inspect energy losses in the building. 3D modeling is also useful in various robotics applications where motion planning in 3D is necessary. A lot of work has been done in the area of environment exploration Ekman et al. (1997); Ðakulović et al. (2011); Yamauchi (1997); González-Baños and Latombe (2001) where the aim is to get a model of environment (either 2D or 3D) using only 2D information captured from the certain height level. Surmann et al. (2003) take 3D scan based on 2D map generated during the exploration. These methods can be used for 3D exploration in simple envi- ronments or when the dense model is not necessary. The obtained model would probably contain also unexplored volumes caused by the assumption that 2D exploration provides 3D coverage of the environment. Blaer and Allen (2007) present an 3D approach method for large outdoor environments in which the 2D map of the environment is needed in advance. After exploration based on 2D map ⋆ This research has been supported by the European Community’s Seventh Framework Programme under grant No. 285939 (ACROSS). the process continues with the 3D exploration. The typ- ical runtime between two scans was about 15 minutes. Dornhege and Kleiner (2011) use a frontier based method extended to 3D exploration. This method requires high computational effort and operating environment is limited to small workspaces. Shen et al. (2012) proposed a stochas- tic differential equation-based exploration algorithm to enable exploration in 3D with limited onboard sensing and processing constraints for micro-aerial vehicle. The main contribution of this paper is a new exploration strategy that addresses the following challenges of model- ing large indoor environments. To overcome the problem of enormous memory consumption while exploring the en- vironments and to reduce the computation effort the pro- posed exploration strategy divides the environment into enclosed spaces and explores until the whole environment is covered. While exploring a room only a local map of the room is in use and computational effort depends only on the size of currently explored room instead on the size of the explored environment as with existing 3D exploration strategies. Furthermore, the proposed strategy prevents the robot to jump between far away scanning positions since the exploration stays inside the detected enclosed space until it is fully modeled. The concept of the proposed exploration strategy is shown in Fig. 1. The exploration starts with the empty map of the environment. In the beginning the robot explores 3D environment using a 2D based exploration algorithm (Section 2). When the robot detects a room the algorithm switches to 3D based exploration inside detected room. For that purpose, we have developed a new room detection algorithm (Section III). Based on 3D exploration method (Section 4) the room is being explored until the whole space inside the room is captured by the 3D sensor and the exploration process is switched back to 2D based Preprints of the 19th World Congress The International Federation of Automatic Control Cape Town, South Africa. August 24-29, 2014 Copyright © 2014 IFAC 10188