Real-Time Indoor Mapping for Mobile Robots with Limited Sensing Ying Zhang, Juan Liu Gabriel Hoffmann Palo Alto Research Center Palo Alto, California 94304 Email: yzhang@parc.com Mark Quilling, Kenneth Payne Prasanta Bose, Andrew Zimdars Lockheed Martin Space Systems Center Sunnyvale, California 94089 Email: mark.l.quilling@lmco.com Abstract—Mapping and localization for indoor robotic nav- igation is a well-studied field. However, existing work largely relies on long range perceptive sensors in addition to the robot’s odometry, and little has been done with short range sensors. In this paper, we propose a method for real-time indoor mapping using only short range sensors such as bumpers and/or wall sen- sors that enable wall-following. The method uses odometry data from the robot’s wall-following trajectory, together with readings from bumpers and wall sensors. The method first performs trace segmentation by fitting line segments to the noisy trajectory. Given the assumption of approximately rectilinear structure in the floor plans, typical for most indoor environments, a probabilistic rectification process is then applied to the segmented traces to obtain the orthogonal wall outlines. Both segmentation and rectification are performed on-line onboard the robot during its navigation through the environment. The resulting map is a set of line segments that represents the wall outline. The method has been tested in office buildings. Experimental results have shown that the method is robust to noisy odometry and non-rectilinear obstacles along the walls. I. I NTRODUCTION A. Motivation During the last decade, research on robotic mapping and localization for exploration and navigation in indoor envi- ronments has made significant progress. Most existing work has used robots with a comprehensive sensor suite, such as ultrasonic rangers, LIDAR, and cameras, and computers with substantial capabilities. This perceptive sensing capa- bility, together with the robot’s odometry, can be used for simultaneous localization and mapping (SLAM), using proba- bilistic or constrained optimization approaches [8]. While such approaches have been popular and successful, they typically require sensor-rich robots and powerful computation, which may be infeasible for many low-power robotic systems. In this paper, we address a quite different mapping problem, using only odometry data and short range sensors, and with suffi- cient algorithmic efficiency to run on low-power embedded processors in real time. The robot platform for this work is the iRobot Create, a commercially available, cheap testbed. The built-in sensing capability of these robots is rather primitive, consisting of: • odometry, measuring distance traveled and angle turned; • a wall sensor on the right side of the robot, sensing proximity to the wall; and • bumper sensors on the left and right of the front side, sensing contact with obstacles. Note that wall and bumper sensors are both short-range sensors; the wall sensor has a typical range of under 10 cm. There is low correlation between measurements over even short distances of travel, hence it is challenging to build a map with these “blindfolded” robots. For this robot platform, we use a wall-following behavior for exploration and navigation to enable the robot to travel around the boundary of a region (see an example floor plan in Fig. 1 (b)). Our algorithms are designed to obtain wall segments for representing an outline of a map. For many applications, such an outline map is useful for identifying corners and intersections which are typically the most critical places in robot placement for surveillance or networking. For a robot exploring by wall-following, it is also important to know if it is looping around an internal island, or it has already traversed the whole external boundary. What we will focus in this paper are efficient algorithms to extract an outline map on-board and in real-time, using noisy odometry and limited sensor readings. Loop detection and closure, due to space limitations, will be presented in another paper. B. Overview Due to the lack of long-range perceptive sensing, we use odometry to estimate wall locations while executing a wall- following behavior. Due to noisy odometry measurements from quantization and wheel slip, the trajectory estimation error accumulates (see Fig. 2 (a)). The goal is to obtain a map from the raw trajectory data. To correct noise in odometry data, we take advantage of prior knowledge regarding the navigation environment. In this paper, we assume an indoor environment, mostly consisting of straight walls and rectilinear corners. The straight wall and rectilinear turn assumptions are strong enough to eliminate odometry noise. This can be considered as a rectification process, where raw odometry angles are classified into discrete directions (multiples of π/2). On the other hand, even in indoor environments we often observe non-rectilinear angles due to obstacles such as small furniture or chairs. Straight- forward rectification would fail in this case. To enable mapping in the presence of obstacles, we have designed a probabilistic