Design of Map Decomposition and Wiimote-based Localization for Vacuuming Robots

Transcription

1 Design of Map Decomposition and Wiimote-based Localization for Vacuuming Robots Cia-Pin Kuo, Sy-In Hwang Department of Computer Science and Engineering Yuan Ze University Abstract-Vacuuming robots ave been more and more popular in te ouse cleaning. But most of suc products on te market do not work efficiently. Te common drawbacks are random cleaning pat, lack of map information, and excessive power consumption. To solve te problems, a localization system is needed. Te tecniques suc as Zigbee, wireless or ultrasound were adopted in te literature. However, tese metods suffer from ig setup cost and large position error. Tis paper proposes a creative projection localization system equipped wit a Wiimote controller and a digital compass. In addition, we apply a Boustropedon algoritm (Back and fort motion), and a map decomposition algoritm to determine a better cleaning pat. By implementing te robotic system, we not only improve te vacuuming robot s efficiency but also prove te practical use in robotic localization. Keyword: Vacuuming Robot, Wiimote, Boustropedon, Digital Compass, Map Decomposition. 1. Introduction Intelligent robot developments ave become more and more popular in recent years. Many robotic researcers are trying to create smarter robots and let tem elping people in our life. Te famous one is ouse cleaning robot. For example, tere are irobot s Roomba, Electrolux s Trilobite, Yujin Robotics i-clebo, and LG s V-R4000, etc. Common features tey ave are sceduled to clean periodically, dodge obstacles, auto-recarge, and detect floors. If people ave busy works every day, vacuuming robot must be a good elper. Unfortunately, tere are some drawbacks wen te robot works. Take Roomba for example, one of tese is it cannot plan cleaning pat. It starts wit spiral circles and ten takes a random direction to clean. On tis circumstance, te robot is likely to waste time cleaning te same areas or miss some regions. In addition, excessive power consumptions may be appened. Te paper will describe a low cost, accurate, and creative localization solution to assist vacuuming robot in te ousecleaning. Taking costs and programmability as considerations; we coose irobot s Roomba for te robot platform. Te main idea is using Wiimote s camera to capture IR LED set up on te robot and support position data. More details will be described in te 3rd sections below. 1

2 2. Related Work Indoor localization must be precision and more stable tan outdoor system. But nowadays systems error ranges are still about fewer meters. Te common tecniques are suc as follows: ultrasound [4], infrared (IR) [1], wireless [5], Zigbee [7], and RFID [3]. Expensive set up costs are teir current features. In te map system, CMU s Coset ad proposed a cellular decomposition [2] tecnique and let te robot take Boustropedon motion in te decomposed cell; Zack J. Butler also yielded a rectilinear decomposition [8] for robot working in special platform. Automatic projector calibration is a popular application. Troug calibrating steps can gain more accurate coordinates. Tis kind of tecnique is useful in electronic wite board, multi-touc, and oter multi-media programs. Raul Suktankar proposed camera-based computer vision tecniques [6] for surface projecting. Te metod is using te omograpy matrix to correct te different images. 3. System Overview Our system includes robot controlling modules and some applications, as sown in Figure 1, te arcitecture below demonstrate VIPER embedded board connects Roomba and compass troug I/O ports. 3.1 Hardware introduction Wole system bundles many components no matter ardwire or software. In te brginning, we coose te Roomba Discovery as a main platform, because it not only offers SCI document for programming but also embeds many functions, suc as bounding cleaning, virtual wall, and auto-recarge. Second, te VIPER embedded board, as sown in Figure 2; all programs will be cross-compiled and downloaded on it. Figure 2. VIPER system board Te next is digital compass wic can assist robot to correct its own directions. In order to gain te compass data easily, we use Basic Stamp 2 (BS2) IC to connect wit it. Te Figure 3 below sows te Parallax BS2 IC combines te Hitaci HM55B digital compass. Figure 3. BS2 IC and digital compass Figure 1. System arcitecture Te finally is Wiimote game comptroller. It is produced by Nintendo Company Ltd. Embedding 3-axis accelerator cip and IR camera inside. For te reason, Wiimote can sense user s swing motion and control game program. Our paper will use its IR camera to capture IR 2

3 LED sources equipped on te Roomba, as sown in Figure 4; te IR camera (Red circle) is in front of te controller. Depending on tis tracing relationsip, we can know were te robot is easily. Te Figure 5 s mat formation is: p ' = Hp x' => = y' 1 => x' = x + 12 y + x + y + 32 x y 1 y' = x + y , 31x + 32 y + 33 Figure 4. Wiimote and IR sensor bar 3.2 Development software In order to implement our system, we also need some application tools to build tem. For example, VIPER embedded board offer its cross-compiler environment named arm-linux-gcc 3.4.x. So we will use tis tool to burn our C code into te board. In oter ands, Roomba and Wiimote release teir free API for programming, just like Serial communication interface (SCI) and Libwiimote-0.4. Finally, wen analyzing Bluetoot packets, System must be installed Hidp, Bluez, and Bluez-libs-devel tese packages on it. In order to figure out te matrix value, we need at least four points in te plane. Toug points calibration, Wiimote camera will map te IR pixels to te projection ones. If te projector is flipped upside down or rotated, te omograpy matrix will correct te target surface pixels automatically. Wen omograpy process is completed, projection pixels must be translated to te world distances. Because slopping projection will lead to te trapezoidal image, as sown in Figure 6, ere we will use bilinear interpolation to solve te geometric problem. 4. Using te Homograpy Te Wiimote-based localization system will use image omograpy process. Te main idea is to find out te omograpy matrix between source image and target plane, as sown in Figure 5, te matematics matrix can warp te IR sources pixels to fit te projection plane. Figure 5. Homograpy projection Figure 6. Geometric transformation Te transformer formation is on te following: x '= a + a x + a y + a xy y ' = b + b x + b y + b xy 5. Map Creation and Pat Planning Our paper ere will introduce a map system. How does it store te data, create, decompose and merge for eac oter? In addition, te vacuuming robot can take some pat strategies to fit te different situations. All of te details will be described below. 3

4 5.1 Rectilinear map In order to demonstrate te robot s back and fort motion clearly. We coose te rectilinear map to implement. Assuming tat map and obstacles outline are all rectangles. System will record objects top-left and bottom-rigt coordinates data, as sown in Figure 7. called target cell. Tey can be seen in Figure 9, te wite spaces are te missing ones Roomba can not pass. Figure 9. Te Final cleaning cells Figure 7. Recliner map expression Map decomposition Our map system will decompose and merge by some rectangle obstacles. System first finds out blocks forward and backward edges and ten cuts along wit tese sides. Now we can get raw regions called cell, as sown in Figure 8, map is decomposed for ten cells by A, B, and C blocks. 5.2 Boustropedon motion Te Boustropedon pat is most likely te cow sowing in back and fort motion, as demonstrated in Figure 10. Te pat s feature is simple and no repeated. For tis reason, robot can clean faster and more efficiently. In order to let te Roomba walk like tat, we also need te digital compass to correct its forward direction. Figure 10. Boustropedon motion Figure 8. Decomposed cleaning cells Ten tese cells are able to searc for eac oter and ceck if need to be merged. If te cell widt is less tan te Roomba s size, suc as cell 1, 4, 7, and 10 above, tey will be combined by te neigboring bigger cells. Wen te map creation process is completed, system will generate final cells for cleaning. Tese cells are 5.3 Pat planning Ones robot as received working instructions; it will start to go te target cell and clean. But in oter situations, robot may take different strategies to fit tem for example contacts obstacles wen exploring To find te starting point If te Roomba confirm were te cleaning area is, it will go to one of te nearest corners for te cell and ten stand by. We can classify te follow situation: in or out te cell, as sown 4

5 in Figure 11, robot first will walk by y axis and ten x axis. Tis section ere will describe te implementation and results. 6.1 Environment introduction Te Wiimote is set up in te ig stick tall about 180 cm. Te front camera is face to te ground sloppily wit 60 degrees angle, as sown in Figure 14. Figure 11. out of te cell (left); in te cell (rigt) Dodging planning It is possible tat robot comes across obstacles wen cleaning. Terefore it take dodging strategy to walk along by te blocks wit position assistance. Robot will compare its original location wit target point, and ten cose te sortest pat to dodge along, as sown in Figure 12. Figure 14. Wiimote controller setting up Finally, te testing range robot will clean is about 205 x 160 (cm), wole localization environment is demonstrated in Figure 15. Figure 15. System overview Figure 12. Roomba dodges along block B Wen te robot is in te target cell successfully, it will continue to clean in Boustropedon motion, as sown in Figure Motion testing Te map s scenario is wit two blocks in it. Te robot s dodging and Boustropedon pat is as Figure 16 (a) (b) sown. (a) Dodging (b) Boustropedon Figure 16. Pat testing; (a), (b) Figure 13. Boustropedon in te target area 6. Implementation All of te above testing videos are on te following links: a. ttp:// 5

6 demo_1.rar b. ttp:// demo_2.rar 7. Conclusion Our paper is successful in implementing an accurate localization system wit a Wiimote game controller. After geometric transformation, te error range can be limited in as small as 2 cm. Furtermore, wile a Wiimote controller is ceape (less tan one tousand NT dollars) and easy to andle, everyone can use it at ome conveniently. To summary, our researc acieves te following goals: Using an embedded board, and some sensor devices to control te popular Roomba robot successfully. Integrating a Wiimote controller to facilitate te calculation of position data. Designing map decomposition algoritm to improve te cleaning pat. Implementing te Boustropedon motion to let te robot move in a regular back-and-fort pat and using digital compass to correct its moving directions. Vacuuming robots can adopt some pat strategies to dodge obstacles. Our system can still be improved in te following directions: Using gyro sensor to replace te digital compass to avoid unnecessary magnetic disturbance. Implementing a multi-robot system for better cooperation. Improving te map system to fit te real environments in our life. Reference [1] A. Harter and A. Hopper, A distributed location system for te active office, IEEE Network, vol. 8, no.1 January/February 1994, pp [2] Howie Coset, Coverage of Known Spaces: Te Boustropedon Cellular Decomposition, IEEE Autonomous Robots, Vol.9, No.3, pp , [3] Lionel M. Ni, Yunao Liu, Yiu Co Lau and Abisek P. Patil, LANDMARC: Indoor Location Sensing Using Active RFID, First IEEE International Conference on Pervasive Computing and Communications (PerCom'03) Marc 23-26, [4] Nissanka B. Priyanta, Anit Cakraborty and Hari Balakrisnan, Te Cricket Location-Support system, Proc. 6t ACM MOBICOM, Boston, MA, August [5] P. Bal and V. Padmanaban, RADAR: An In-Building RF-Based User Location and Tracking System, Proc. IEEE INFOCOMM 2000, Mar [6] Raul Suktankar, Robert G. Stockton, Mattew D. Mullin, Smarter Presentations: Exploiting Homograpy in Camera-Projector Systems, Proceedings of International Conference on Computer Vision, [7] Stefan Scwarzer, Martin Vossiek, Markus Picler and Andreas Stelzer, Precise Distance Measurement wit IEEE (ZigBee) Devices, Radio and Wireless Symposium, 2008 IEEE. [8] Zack J. Butler, Contact sensor-based Coverage of Rectilinear Environments, Proceedings of te 1999 IEEE International Symposium on. 6