Electrical Engineering in Japan, Vol. 190, No. 4, 2015 Translated from Denki Gakkai Ronbunshi, Vol. 133-D, No. 5, May 2013, pp. 502–509 Control of Autonomous Mobile Robot Using Map Matching with Optimized Search Range KOUHEI KOMIYA, 1 SHUNSUKE MIYASHITA, 2 YUTAKA MARUOKA, 1 and YUTAKA UCHIMURA 1 1 Shibaura Institute of Technology, Tokyo, Japan 2 Kubota Corporation, Japan SUMMARY Accurate estimation of self-position is indispens- able for autonomous mobile robots. Map-matching-based self-positioning is a promising method; however, adequate matching area selection is an issue to be solved. In this paper, we propose an adjustable map matching method that remedies a defect of the conventional template match- ing technology. The proposed method was evaluated ex- perimentally, and the results confrmed the advantages of the proposed method. C⃝ 2014 Wiley Periodicals, Inc. Electr Eng Jpn, 190(4): 66–75, 2015; Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/eej.22608 Key words: mobile robot; self-localization; particle flter; map matching; occupancy grid map; odometry. 1. Introduction With recent progress in automation and instrumen- tation technologies, autonomous mobile robots such as automated guided vehicles (AGVs) are widely used in factories and other production facilities. In addition, there has been extensive R&D on robots that can coexist with people in human living environments such as homes, hos- pitals, and offces [1]. In the automation of mobile robots, position control has in many cases been implemented using guidance by magnetic tape, and so on; however, from the standpoint of equipment cost and maintenance, advanced autonomous mobile technologies are needed to made possible motion without any additional track equipment [2]. To arrive reliably at a destination, autonomous mobile robots must recognize the surrounding environment and accurately estimate the self-state (position and orientation). Dead reckoning is widely used for self-localization of autonomous mobile robots, employing information ac- quired by wheel encoders, inertial sensors, or other inter- nal sensors. However, in dead reckoning and other self- localization methods based on the integration of measured values from internal sensors, accuracy may be degraded by cumulative errors. Therefore, external sensors must be employed for self-localization when internal sensors alone cannot assure suffcient accuracy. Self-position measurement using GPS as an external sensor has gained popularity, but there are problems such as errors in handling multipath transmission [3]. In addition self-localization methods using geometric information on the surrounding environment acquired by an laser range fnder (LRF) have been proposed. The LRF is an active sensor that illuminates objects with laser light, which has the advantage that the infuence of ambient light and other disturbances is weak. In particular, an LRF-based method (map matching) for self-position estimation by matching prepared map data and scan data at the current position has been developed [4]. There are a number of map matching techniques, such as geometric shape matching based on the ISP algorithm [5, 6] and template matching with scan data used as a template and map data used as input [7–9]. The former techniques deal with point cloud data, which allows relatively easy implementation without the need for data processing. The disadvantages are high sensi- tivity to outliers and noise, and computational complexity. On the other hand, the latter techniques have shorter pro- cessing time and higher immunity to noise, but accurate es- timation results cannot be obtained unless the search range is set properly. Aiming at robust autonomous motion of mobile robots, we propose here a multistage map matching method that solves the problems of conventional template C⃝ 2014 Wiley Periodicals, Inc. 66