Target Tracking and Obstacle Avoidance for Mobile Robots

Transcription

1 Target Tracking and Obstacle Avoidance for Mobile Robots Ratchatin Chancharoen, Viboon Sangveraphunsiri, Thammanoon Navaknlsirinart, Wasan Thanawittayakorn, Wasin Bnonsanongsupa, and Apichaya Meesaplak, Mechanical Engineering Department, Chnlalongkom University, Bangkok, Thailand fmercck2eng.chula.ac.th Abstract The paper presents a real time motion planning and obstacle avoidance for an autonomous mobile robot. The robot is equipped with low-resolution optical sensors and electronic compass and is driven by stepper motor. Thus, there are three guidance modes: target tracking using optical sensors, directional guidance using compass, and dead reckoning. There are another optical sensors equipped on board to detect obstacles. The vehicle is controlled based on the information from these sensors. In the proposed technique, the control algorithm is switched to wall following mode when facing an obstacle. This technique is very simple but efficient. Several simulation and experiments demonstrate good performance even though using low-resolution sensors. Keywords: Mobile Robot, Obstacle Avoidance, Target Tracking, Wall Tracking. 1. Introduction The major advantage of a mobile robot is its unlimited workspace because it has an ability to travel. An industrial mobile robot such as AGV (Automated Guide Vehicle) is considered the most flexible type of material handling system and is normally operated in a large working area where is difficult to control to suit the robot. The robot may encounter unknown environments and/or unexpected obstacles. Likewise, the exploratory or field robot will encounter unknown environment and/or unexpected obstacles during its normal operation. Thus, the major challenge IS how the robot can deal with unknown environments andor unexpected obstacles efficiently. There are a number of researches in this area in the past twenty years. The brief history is discussed in the following section. 2. Previous work The most simplistic reactive collision avoidance for an autonomous mobile robot is called Wander [I]. Everett [2] has implemented this technique to his ROBART mobile robot series, including ROBART I ( ), ROBART I1 ( ) and ROBART 111 (1992). ROBART 111 is intended to be an advance demonstration platform for non-lethal response measures incorporating the reflexive teleoperated control concepts developed on ROBART 11. The concept of potential fields was introduced by Krogh [3] in 1984 for simulations of localized mobile robot control, and by Khatih [4] in 1985 for manipulator control using Cartesian as opposed to joint coordinates. Zelinsky has applied the concept of potential fields to his NOMAD2000 [5]. In 1985, the Certainty Grid algorithm was presented by Moravec and Elfes [6]. This algorithm was described as a scheme for mapping imprecise sonar range returns into certainty grid using probability distribution functions. In 1990, Borenstein and Koren [7] developed the Vector Field Histogram (VFH) technique. This technique is a combination of both Potential Field and Certainty Grid. The VFH algorithm was initially implemented on a modified cybermotion K2A platform equipped with an onboard based PC-compatible computer. Since then, several techniques based on VFH have been proposed including DVFH (Double Vector Field Histogram) by Hong Yang et al. [SI, and VFH+ and VFH* by Iwan Ulrich and Johann Borenstein [9]. The applications of obstacle avoidance techniques can be seen in Navbelt and Navchair. Shraga Shoval et al. have implemented the concept of obstacle avoidance into Navhelt [IO], which is a new traveling aid for the blind in Johann Borenstein introduces Navchair [I I] (An assistive Navigation system for Wheelchairs Based upon Mobile Robot Obstacle Avoidance) which using Vector Field Histogram. In 2001, Giunluca Antonelli et al. [12] presented Navigation and Guidance System (NGS) for autonomous underwater vehicle. The target tracking and obstacle avoidance have already been implemented in several commercial products. SONY has introduced AIBO [13], which is the robot that has an ability to avoid obstacles. Activ Media Robotics [I41 has launched AmigoBOTTM, which can wander around the house, avoiding obstacles detected by its sonar. Kevin Hakala [15] has built a lawnmower, called LawnNibbler. It can mow and avoid tree, pets or children in its area. For the industrial products, Activ Media Robotics [I61 has introduced PowerBotTM robots as a high-payload intelligent mobile platform that can circumvent obstacles and still reach its destination. International Submarine Engineering (ISE) Ltd [I71 has a Strategic Technologies for Automation and Robotics (STEAR) project to develop autonomous controlled robotic manipulators, which have the ability to plan obstacle free trajectories throughout unstructured environments. In this project, an intelligent mobile robot has been developed with three modes of path planning: target tracking, direction guiding, and dead reckoning. The /021$17.00Q 2002 IEEE. 13 IEEE ICIT 02, Bangkok, THAILAND

2 paper addresses the motion planning and obstacle avoidance for this robot. The mobile robot is to track the light as seen in Figure 1 while avoiding collision with the wall. Detail of the robot and algorithm is explained in the following section. Several simulations and experiments are used to demonstrate the success of the algorithm. the distance between the robot and the obstacle. If the obstacle is sufficiently near, then the wall following is active. This routine is very simple. Details of The Tracking method and the wall following method are explained next. The Tracking method In order to control the robot to track the target, we uses signal from three LDR tracking sensors to generate the command to the robot wheel. Consider when signals from two tracking sensors in front of the vehicle are stronger that the sensor at the back. The tracking sensors are 8 bits resolution. The signals from these sensors at the front are compared, and then the signal difference is partitioned into 11 levels. The motor command is then determined as shown in Table 1. This method balances the signals from both sensors. Ta Figure 1 The intelligent mobile robot and a light bulb as a target. 3. The robot platform and algorithm The mobile robot for this research is shown in Figure 2. It has been constructed from a commercial Rcbo- BASIC' mobile robot. We have installed electronic compass to guide the robot, three LDR optical sensors as target tracking sensors, two optical sensors as line tracking sensors, which pointing to the ground and capable of determining the its color, and four optical sensors around the robot as wall following sensors. With these modifications, the mobile robot has tbree modes of guidance: target tracking, direction guiding, and dead reckoning. In the first mode, the robot tracks the target using the signals from three LDR opt.ical sensors. If the signals from both two LDR op1:ical sensors in front of the robot are balanced, the head oi'the robot is in the direction to the target. In the second mode, the direction guidance uses electronic compass to command the robot direction. The compass informs the robot orientation via serial communication. In the third mode, the dead reckoning is obtained by open 1.00~ control of stepper motors. ' The wall following method In the wall following mode, we uses two of the four wall-tracking sensors to generate the motor command. The wall following sensors are proximity sensors which have only three states: short (S), mid (M), and far (F). In case that the wall is on the right of the robot, hvo sensors on the right will he used. If the signals from both sensors are balance, the robot is in parallel with the wall. However we need to control the distance to the wall as well. We experimentally determine the appropriate motor command to control the robot orientation as well as the distance to the wall as shown in the following table. [-qd "%- I i o, X2&zm- Ta Figure 2 The modified Robo-BASIC@ mobile robot. In the proposed technique, the robot switches between two modes: tracking and wall following, dependin: on 14 IEEE ICIT'02, Bangkok, THAILAND

3 . Simulations and results Series of simulations are used to demonstrate the performance of the wall-following algorithm. In the first simulation, the robot is to track a target at various positions. The sensors onboard are able to determine orientation and distance of the target relative to the vehicle. Result, shown in Figure 3 demonstrates that the robot can reach all targets along a smooth trajectory. If the obstacle is in a circular shape as shown in Figure 5, the proposed technique is still able to complete the task. The robot is able to maintain the distance to the wall when operated in the wall following mode. - P A T ~ ~ ~ C T modi ~ W I A. o~.f.~i~~ab.nn ~~i~ Figure 6 The trajectory when tracking a target in maze. Figure 3 Trajectories of robot tracking target In the second simulation, there is an obstacle blocking a direct path to the goal as seen in Figure 4. The range finders equipped onhoard are able to detect this obstacle. When the vehicle approached the obstacle, the wall following mode is then active. The vehicle redirects its path and adjusts its orientation to he in parallel with the wall and thus, preventing the collision with the wall. The wall following mode is deactivated when the vehicle finds a new direct path to the goal. 1L/ I iim &I A i Consider the next simulation when the robot is in maze while tracking the target (Figure 6). The proposed technique successfully drives the robot to the goal and avoids collision with complicated maze. In the next simulation, the forbidden area is presented as shown with black line in Figure 7. The robot distinguishes the areas by using two sensors pointing to the ground. When facing the forbidden area, the robot tries to keep one sensor on the forbidden area and the other on the free area. The wil1 keep the robot to be in parallel with the area boundary. The result demonstrates that the proposed technique is able to drive robot to the goal and not enter the forhidden zone.?- 4 7'1 Figure 7 The trajectory when facing forbidden area. Figure 8 shows that the robot is still reaching to the goal even though both obstacle and forbidden area are presented. The simulation demonstrates that the proposed technique is simple hut efficient. In the next simulation (Figure 9), the mobile robot uses electronic compass to guide the vehicle direction. The robot is programmed to go north except when facing with an obstacle. This simulation demonstrates that the wall-following algorithm can be used in conjunction with direction guidance. 15 IEEE ICIT'02, Bangkok, THAILAND

4 - O L a A T-IT~UUM& Aor~rru~.d-.~~& Figure 8 Both forbidden area and obstacle are presented. In the following tests, the robot is to track the light bulb and avoid the collision with wall and not enter the black zone. The marker pen is attached to the robot to create the trajectory (Figure 11). I In the first experiment, the robot is to track the target at various positions. Figure 12 shows that the robot can reach the goal for all cases. The robot tries to balance the signals from both optical sensors in front of the robot and thus its head is facing the light bulb. This technique results in a smooth trajectory, which is similar to the simulation result in Figure 3. l QC- 0 Yn & Inn xm?d. A= A TWT-U.* b o~-u+.,.ib~.mou Figure 9 The robot is heading north and facing a wall. Figure 12 bulb) O L X Figure 10 The trajectory when facing a moving obstacle. The wall following algorithm is also tested with a moving obstacle. The result is shown in Figure IO. The robot is heading to the target directly until it encouiters the obstacle as marked 4 in the picture. The wall following mode is then active and keep the orientation of the robot to be in parallel with the obstacle. When the obstacle is not blocking a direct path to the goal as marked 7, the wall following mode is deactivated. The robot adjusts its orientation and moves to the target. 5. Experiments and results After we have performed a number of simulations, we implemented the motion planning & obstacle avoidance to the intelligent mobile robot. The mobile robot using in the experiment has the same dimension as the one using in the simulations. We have experimentally tested the wall following algorithm as discussed next. Figure 13a and 13b show the experimental results when the forbidden (black) area is presented. The robot uses optical sensors at the bottom to detect the forbidden area. When this area is detected, the robot tries to keep one sensor on the forbidden area and the other on the free zone. This keeps the robot to be in parallel with the area boundary, and thus prevents robot to enter the forbidden zone. We have tested with various initial robot positions and various target positions. Figure 13c shows the result when an obstacle is presented and the goal is at various positions. The wall following algorithm has been tested with direction guidance using electronic compass. The robot is programmed to go north and not to enter the black area. Experimental results show that the robot can reach all the targets (Figure 13d). The robot is also tested in the area where an obstacle is presented. Experimental results in Figure 13e shows that the robot can escape the wall and go north. 16 IEEE ICIT 02, Bangkok, THAILAND

5 ~~~~ application a) Various initial positions. LM *c4 [I] H.R.Everett, Sensor for mobile robots: Theory and (A K Peters, Ltd, 1995). [2] [3] B. H. Krogh, A Generalized Potential Field Approach to Obstacle Avoidance Control: International Robotics Research Conference, 1, Bethlehem, Pennsylvania, August, c) Various target positions and an obstacle is presented. xm."j!l./- ~ 0 tod me se4 800 X W d) The robot is heading north and facing a forbidden zone. e) The robot is heading north and facing an obstacle Figure 13 The trajectories for various cases. These experiments demonstrate that the wall following algorithm is simple hut efficient. The robot can reach the.goal with good stability for all test cases. The wall following algorithm can be used with various motion planning techniques and its implementation is easy. 6. Conclusion The simple wall following can be used as collision avoidance technique. This technique can he integrated with various guidance modes such as target tracking, direction guidance, and dead reckoning. When approaching an obstacle or facing forbidden area, the robot will hy to maintain the distance to the obstacle while tracking the object. This prevents the robot to make a collision with the obstacle. 7. Acknowledgements The authors gratefully acknowledge the grant support from MTEC (National Metal and Materials Technology Center). Robotics Laboratory, Dep. Of Computor Science, University of Wollongong, Australia. [6] H.P. Moravec, A. Elf&, High Resolution Maps from Wide Angle Sonar, Proceeding of the 1985 IEEE International Conference on Robotics and Automation, St. Louis, MO, pp , March, I [7] J.Borenstien and Y.Koren, Real-time Obstacle Avoidance for Fast Mobile Robots in Cluttered Environments, Proceedings of the 1990 IEEE International Conference on Robotics and Automation, Cincinnat, Ohioi, May 13-18, 1990, pp [8] H. Yang, J. Borenstien, D. Wehe, Sonar-based Obstacle Avoidance for a Large,Non-point,Omidirectional Mobile Robot Deoartment Of Mechanical Engineering, University of Michigan, USA. [9] I. Ulrich, and J. Borenstein, VFH*: Local Obstacle Avoidance with Look-ahead Verification, *L A&.m.eo >A-' Proceedings - of, the 2000 IEEE International Conference on Robotics and Automation, San Francisco, CA, April 2000, pp [IO] S. Shoval, J. Borenstein, and Y Koren, Mobile Robot Obstacle Avoidance in a computerized Travel Aid for the Blind, Proceedings of the 1994 IEEE Robotics and Automation Conference, San Diego, California, May 8-13,1994, pp [ [12]G. Antonelli, S. Chiaverini, R. Finotello, and R. Schiavon, Real-time path planning and Obstcle avoidance for RAIS - an autonomous underwater vehicle, IEEE JOURNAL OF OCEANIC ENGINEERING, Vol. 26, No. 2, April [I31 [ [ [I 61 [ IEEE ICIT'OZ, Bangkok, THAILAND