Data Acquisition of Obstacle Shapes for Fish Robots

Size: px

Start display at page:

Download "Data Acquisition of Obstacle Shapes for Fish Robots"

Leslie Williams
6 years ago
Views:

1 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 Data Acquisition of Obstacle hapes for Fish obots EUNG Y. NA, DAEJUNG HIN, JIN Y. KIM, EONG-JOON BAEK Dept. of Electronics and Coputer Engineering Chonna National University Yongbong-dong, Buk-gu, Gwangju OUTH KOEA Abstract: - It is iportant to get inforation about obstacles to avoid collision for a natural and sooth oveent for fish robots. In this paper, we propose an I sensor-based easureent syste which gets inforation about obstacles for underwater fish robots. Due to coplexity and space restriction, I distance sensors are installed instead of caeras or sonar systes. Much iportance is given to the nearby obstacles than those in the distance. A scanning I sensor and two fixed I sensors are used in our proposed syste to get the pattern inforation of the obstacles in water. By analyzing the I sensor data, the estiated shapes of obstacles are recognized. Trajectory control for underwater robots should be designed based on the estiated results of the shapes. Our experiental results show that fish robots, which use our proposed sensor syste, ake successful oveent avoiding collision through channels. Key-Words: - Fish obot, Obstacle hapes, Collision Avoidance, I Distance ensor 1 Introduction The ost basic task for fish robots is to detect obstacles and to avoid collision to ake sooth and natural oveents. The fish robots need to detect the obstacles in advance and ove without collision. uccessive collisions were prevented utilizing acceleration data when collision occurred in [1]. Fish robots ake appropriate direction changes to avoid collision based on the I distance sensor data. Analyzing iage data of the target areas to recognize possible obstacles is the basic ethod in the water. Despite advantages of iage systes, such as caeras and sonar, there are any kinds of situations where iage data cannot be properly applied. The coon reasons for this restriction are the liited capabilities of processors, ranges of short distances, transission tie and adverse underwater conditions to get sufficient iage data. For sall-scale fish robots, it is necessary to apply cost-effective and basic sensors such as I distance sensors rather than iage sensor systes. I sensors need sall aount of inforation; thus, a quick and effective detection can be ade. But the difficulties of the use of siple distance sensors are to recognize the shapes of obstacles correctly. In addition to a variety of shape patterns of the obstacles, shift distance and rotation angle of the robot relative to the obstacle ake it hard to estiate shape patterns. In our proposed shape estiating syste, a scanning I sensor at the center with two fixed I sensors at each side is used to get the pattern inforation of the obstacles in water. By analyzing the I sensor data which has difficulties due to shift, rotation, and the speed of the robot, the estiated shapes of obstacles are reconstructed. Path planning or control for fish robots should be designed based on the estiated results of obstacle shapes. Our experiental results show that fish robots, which use our proposed sensor syste, successfully ake their oveent avoiding collision through the various types of channels. 2 I Distance canning yste for Fish obots The fish robot syste in our lab uses I distance sensors to easure the distance fro the robots to obstacles in water. Due to structural liits, a resolution level depends on the distance and is also affected by noise. The underwater environent causes

Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 31 a uch higher level of noise to I distance sensors; oreover, the range of

2 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 31 a uch higher level of noise to I distance sensors; oreover, the range of underwater easureent gets ore shortened than that of the easureent in the air. ince data can be obtained only for one point at one tie, it akes the syste insufficient for shape recognition. To deal with these disadvantages of I sensors for shape recognition, a proper scanning ethod should be used. In this paper we use a scanning I sensor with a servo otor at the center, with two fixed I sensors at each side, to get the pattern inforation of the obstacles in water. Also, a potentioeter is used to easure the rotation angle of the otor. The I distance sensor, which is used in this experient, is a GP2D by harp Co. The range of easureent in the air is -8 c, but the valid range in water is uch reduced to - c due to noise. The noise level becoes higher as the distance gets longer. The device of a sensor odule is treated water-proof to be used in water. A fil is attached to the lenses so that direct exposure to water as well as reduction of effective distance range can be prevented. 3 ecognition of Obstacles The underwater robot which was constructed in our lab is c, length, width, and height, respectively. The robot oves in the tank with the size of 1 1 c. Obstacle odels, which consist of two planes, range fro 9 to 27 degrees in the water tank as shown in Figure 3(a). Fish robots can approach obstacles with different angles relative to the center of obstacles as shown in (b). Fig. 1. A fish robot used in experients (a) plane angles of obstacles Obstacle Position of a fish robot -º º -º º º (b) rotation of obstacle Fig. 3. Various shapes and positions of obstacles Fig. 2. Infrared distance sensor scanner The fixed sensors, which are located on each side of I distance scanner, are used to copensate the distance to the obstacle. They are also used as a priary data for a quick easureent or collision avoidance in the eergencies that no accurate object recognition is available. The shape easureent syste is shown as in Figure. 2. Figure 4 shows a typical exaple of data processing which geoetrically reconstructs an original obstacle shape using angular and distance inforation obtained within the easureent range in static condition.

3 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 32 (a) distance raw data soe sensors do not provide distance data for obstacle shape reconstruction, are as shown in Table 1. Zero and 1 refer to beyond the easureent range and within the range, respectively. We assue that the rules in Table 1 are applied only when all sensor values are less than c fro the center. Table 1. iple avoidance rules of fish robots 5 Case P P P Coands (b) angle raw data C1 Move Forward C2 1 Turn eft C3 1 If α >, Then Turn eft Else If α <, Then Turn ight (c) reconstructed obstacle shape Fig. 4. Typical exaple of an obstacle shape reconstruction in static condition 3.1 iple ecognition When the obstacle is located too close fro the robot or there is not enough tie to obtain sufficient data for the obstacle, a quick and siple ethod is utilized to avoid collision. The location data for the shape of the obstacle is calculated using the absolute coordinate values easured by all sensors. The three points easured by the scanning sensor and two fixed sensors and, which are located on each side, are P, P, and the coordinates are ( 5, + 4), ( x, y ) and ( + 5, + 4), respectively. and are easured distances fro the left and fro the right fixed sensors, respectively, and is the value by the scanning sensor. If the angle of the scanning sensor is α, then x = ( + 4)sinα, y = ( + 4)cosα The easureent frequency of the sensors is Hz. For the scanning sensor it oves sideways by a servo otor with a constant speed within the range of - ~ + degrees. canning frequency is 1/3Hz. The three basic values fro three sensors are obtained alost at the sae tie. ules for collision avoidance in the eergency, when a robot is too close to an obstacle or C4 1 1 C5 1 Turn ight C6 1 1 C7 1 1 C If θ >, Then Turn ight <, Then Turn eft If P > P, Then Turn eft Else If P < P, Then Turn ight If θ >, Then Turn ight <, Then Turn eft When α <, If θ >, Turn ight, <, Turn eft When α >, If θ >, Turn ight, <, Turn eft 3.2 Accurate ecognition ince shapes vary fro obstacle to obstacle, we assue, at first, that the odel consists of two planes. Therefore, the reconstructed obstacles always have two ending points and they have one corner point unless it is single sided. This is geoetrically calculated fro the absolute coordinate values of the reconstructed obstacles. But all the data about the obstacle s structure and approaching angles should be available for a reasonable approxiation to obstacle shapes. Figure 5 shows geoetric relationships of sensor data for an exaple of the obstacle. et the lines passing through P be, P be, P be, and the slopes of the lines be,,, respectively, The estiations of the plane angle of an obstacle and rotation of an obstacle, distance to an obstacle are ade fro the easureents of the three I sensors. The ethods such as log-polar ap or fft which require vast aount of calculation are not appropriate for a icroprocessor located inside a sall fish robot.

4 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 33 While a robot is oving there is a considerable level of noise due to waves, vibration of the actuators or the scanning otor itself. Therefore the discrepancy of the scanning sensor data should be copensated using the distance data of the fixed sensors at both ends. We propose a ethod that transfors the raw shape data fro the scanning sensor to the real obstacle shape irrelevant to the rotation and shift of the obstacle using the distance data of the fixed sensors. shows alost constant - as the otor angle changes fro +11 to -8. We assue this constant angle to be. The line of squares indicates +74 at the otor angle of -11 and then it drops sharply. We assue the average of the first three angles to be. Therefore, the estiated obstacle angle is about 89, and the corner point is easured at the average of -11 and (a) canned data (a) Obstacle without corner angle of object angle of otor (b) Estiated angles fro left and right end points Fig. 6. s of an obstacle fro two left and right points to each scanned point Figure 7 shows reconstructed results when the obstacle which has 1 is rotated in five different ways (b) Obstacle with a corner Fig. 5. Geoetry of sensor syste for real obstacles In Figure 6(a), scanned data for an obstacle which has a 9 degree corner is shown in a dotted line. Estiated angles fro the far left easureent are denoted by a line ade up of squares, and the estiated angles fro the far right easureent are denoted by a line ade up of diaonds, as shown in Figure 6(b). The data denoted in a line of diaonds Fig. 7. econstruction and feature points of the sae angle obstacle in five positions

5 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 34 In Table 2, estiated reconstructions are obtained for the cases such as the distance of c, the zero degree of rotation, and the degrees of the obstacles are 9, 1,,, 2,, and 27, respectively. Also, it shows the estiations, in ters of average and standard deviation, of obstacle angles in the first row, and rotation estiation in the ond row, for the obstacles with each degree given. Table 2. Estiation results for various angles of an obstacle (distance is c, rotation angle is and shift distance is c) Error 9 1 [ ] otation [ ] 2 27 ean std ean std Table 3. Estiation results of 1 obstacle for various rotation angles (axiu distance is c and shift distance is c) Error - - ean [ ] std otation ean [ ] std Table 4. Estiation results of obstacle for various rotation angles (iniu distance is c and shift distance is c) Error - - ean [ ] std otation ean [ ] std Table 3 is the estiation results for various rotation angles when the test obstacle has 1 obstacle angle and axiu distance is c. Table 4 is the estiation results for various rotation angles when the test obstacle has obstacle angle, iniu distance is c. Fig. 8. Mebership functions of and The rules for obstacle avoidance use two slopes, and, and the ebership functions are as shown in Figure 8, and the fuzzy rules are described in Table 5. In Table 5, B, F, and represent Backward, Forward, and top of the left and right otors, respectively. Table 5. Fuzzy rules for obstacle avoidance using and NB N ZEO P PB NB N ZEO P PB eft all eft all eft all eft all ight all (,F) (,F) (,F) (,F) (F,) eft all eft all eft all ight all ight all (,F) (,F) (,F) (F,) (F,) eft Big eft all ight all ight all ight all (B,F) (,F) (F,) (F,) (F,) eft Big ight all ight all ight all ight all (B,F) (F,) (F,) (F,) (F,) Backward ight Big ight Big ight all ight all (B,B) (F,B) (F,B) (F,) (F,) 4 Copensation of obot s Moveent Based on the distance and angle data easured by a scanning sensor when the robot is not in otion, the original shapes of obstacles are reconstructed and their characteristics are analyzed. obots are oving in real situations; thus, the effect of distance changes during a half cycle should be copensated for the priary data obtained in the static condition. ince the fixed sensors have ore reliable data, the average value of each side fixed sensor data is used to copensate the scanned shape data. The copensation begins when the iniu distance during a half cycle easureent becoes between c and c. The real-tie distance data and the reconstructed shapes of obstacles which have different angles are shown in Figure 9. Figure shows a typical exaple of applying the proposed ethod to the underwater robot s passage along the given path in a water tank.

2 4 6 8 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 35 eft sensor 28 ight sensor canning sensor (a) Data of two fixed and a

6 Proceedings of the 2nd WEA International Conference on Dynaical ystes and Control, Bucharest, oania, October -17, 6 35 eft sensor 28 ight sensor canning sensor (a) Data of two fixed and a scanning I sensors raw distance copensated distance fixed I sensors at each side is proposed for fish robots. Using this sensor syste instead of caeras or sonar proves to be effective in getting the shape pattern inforation of the obstacles in water for sall robots. By analyzing the I sensor data which has difficulties due to rotation and the speed of the robot, the estiated shapes of obstacles can be reconstructed. Path control for fish robots should be designed based on the estiated results of obstacle shapes. Our experiental results show that fish robots successfully ake their oveent avoiding collision through the various types of channels (b) Obstacle shapes in case of (a) 28 eft sensor ight sensor canning sensor Acknowledgeents This work was supported by the C-HEC, CNU under grant , and NUI-CEIA, CNU (c) Data of two fixed and a scanning I sensors 28 raw distance copensated distance (d) Obstacle shapes in case of (c) Fig. 9. Distance data and copensation of obstacle shapes for oveents Fig.. Exaple of obstacle avoidance 5 Conclusion A shape estiating I distance sensor syste, which consists of a scanning I sensor at the center with two eferences: [1] Na, eung Y., hin, Daejung, Ki, Jin Y., and Choi, u-il, Collision ecognition and Direction Changes Using Fuzzy ogic for all cale Fish obots by Acceleration ensor Data, FKD 5, NAI 36, 5, pp [2] Borenstein, J. and Koren, Y., Obstacle avoidance with ultrasonic sensors, IEEE Transactions on obotics and Autoation, Vol. 4, Issue 2, 1988, pp [3] Ku, Ching-Heng, Tsai, Wen-Hsiang, Obstacle avoidance for autonoous land vehicle navigation in indoor environents by quadratic classifier, IEEE Transactions on ystes, Man and Cybernetics, Part B, Vol. 29, Issue 3, 1999, pp [4] Antonelli, G., Chiaverini,., Finotello,., and chiavon,., eal-tie path planning and obstacle avoidance for AI: an autonoous underwater vehicle, IEEE Journal of Oceanic Engineering, Vol., Issue 2, 1, pp [5] Petillot, Y., Tena uiz, I. and ane, D.M., Underwater vehicle obstacle avoidance and path planning using a ulti-bea forward looking sonar, IEEE Journal of Oceanic Engineering, Vol., Issue 2, 1, pp. -1. [6] Wai,.-J., Chen, P.-C., Intelligent tracking control for robot anipulator including actuator dynaics via TK-type fuzzy neural network, IEEE Trans. Fuzzy ystes, Vol., 4, pp

Vision Based Mobile Robot Navigation System

International Journal of Control Science and Engineering 2012, 2(4): 83-87 DOI: 10.5923/j.control.20120204.05 Vision Based Mobile Robot Navigation Syste M. Saifizi *, D. Hazry, Rudzuan M.Nor School of