Robot Olympics: Programming Robots to Perform Tasks in the Real World

Transcription

1 Robot Olympics: Programming Robots to Perform Tasks in the Real World Coranne Lipford Faculty of Computer Science Dalhousie University, Canada Raymond Walsh Faculty of Computer Science Dalhousie University, Canada October 7, 2009 Abstract This paper describes the challenges and successes we faced programming a robot to compete in the Robot Olympics. We discuss the three programs we developed to handle the tasks, including designs that were tested and rejected. We present the results of the robot s performance in the Robot Olympics and discuss potential solutions for performing the required tasks more efficiently. Introduction Robots are becoming increasingly popular in every day life. With their increase in popularity and use comes the challenge of programming them to cope with every day situations in an ever-changing environment. The Robot Olympics required the robot to perform three distinct tasks: race along a path made from a single black line, perform a scavenger hunt, and participate in sumo wrestling matches. Participation in each of these events required a dedicated program to optimize the robot s performance. The sensitivity of the sensors and reaction time of Lego Mindstorms NXT robot motors were not found to be consistent. This provided a design challenge and we discuss its effect on our design choices as well as solutions we feel would improve the robot s performance. We present the algorithms for the programs we designed and discuss designs we evaluated and discarded. None of the programs we used in the competition were the ones we had originally discussed creating. However, the programs used resulted in a tie for first place in the 2009 Robot Olympics. The Long Line Race The first task of the Robot Olympics required the robot to race along a solid black line. The course included straight sections, smooth curves and corners no tighter than 90 degrees. The robot was required to have its light sensor pass over the line at least once a second and should it leave the course, was required to find the line within 15 centimetres of where it lost it. The robot that completed the course in the least amount of time was declared the winner. [1] Knowing the robot would begin the course with its light sensor on the black line, our original design was to have the robot move straight forward along the line until the sensor left the line. Upon leaving the line, the robot would stop and search for the black line while rotating 90 degrees to the left. If the black line was found, the robot would continue to follow it. If not, the robot would rotate 180 degrees to the right while searching for the line. The line would have to lie within this 180 degrees of rotation because no corner could be at an angle less than 90 degrees to the race line (determined by the race course rules). Once the line was found, the robot would continue to follow it. Any time the line was lost, the searches 1

2 would continue as described. The more time the robot spends moving straight forward, the less time it should take to complete the course. However, in practice the planned solution did not work. The light sensor was not detecting a narrow enough area to make sufficient progress moving forward in a straight line. If the robot moved at even a slight angle the sensor would detect it had lost the line and would begin searching. The increased time spent searching to re-find the line was much greater than any time saved by having the robot follow the line by moving straight forward. In addition, getting the robot to move in a straight line is not an easy task to accomplish. Even slight variances in the actual speed of the two motors would cause the robot to drive on a slight angle, causing it to leave the line more often. Choosing to optimize the existing program allowed us to dedicate more time to developing our program for the Scavenger Hunt event. The program ultimately used to compete in The Long Line Race was based on the line following program in the course notes (page xxx). We felt that the program did an adequate job following the line, and with minor modifications to the power could do an even better job. The program algorithm is as follows: Start: Robot wheels on the provided tape starting marks (light sensor on black line to be followed) Steps: Loop Infinitely: Move forward with wheel C Wait for light sensor to detect white wheel C Move forward with wheel B Set timer to 0 and start it While (dark line not detected) IF (timer > 1) Move backwards with wheel B Wait for light sensor to detect black line Move forward with wheel B Reset timer To test for optimal speed we set up the following sample course: Figure 1. The Long Line Race Test Course 2

3 We tested the program on the course at four different powers with the following completion times: Power Test One Test Two Average sec 38 sec 43 sec sec 45 sec 48 sec sec 33 sec 32.5 sec sec 35 sec 32.5 sec Figure 2. The Long Line Race power test results. We ultimately decided to use 70 power for the motors. While the average speed completion time for 75 power and 70 power was the same, the robot made smoother progress and had the fastest single completion time at 70 power. The majority of the race time was spent on the tight curve at the top of the course. Because the robot made smoother progress through the gentle curves, corners, and straightaways at 70 power and we felt the majority of the course would be made of those components, we chose 70 power as the speed for the motors. We also discovered early in the testing phase that the light sensor needed to be close to the wheels of the robot for this program to work efficiently. The farther the sensor was from the wheels, the more difficulty the robot had finding the line on the turns, especially tight ones. This required the light sensor to be moved from its Scavenger Hunt position when competing in The Long Line Race. During testing the robot completed the course every time. The tight turn at the top of the course required the robot to do the most searching for the line but did not prevent it from completing the course. At the competition, even though the robot had difficulty with the series of tight turns (similar to that at the top of our test course), it completed the course on all three attempts and was one of the few to do so. Sumo Wrestling The second Robot Olympics event was a series of Sumo Wrestling matches between two robots. The Sumo ring was two concentric squares approximately 0.5 metres apart and demarcated by black tape. The robots began in opposite corners of the inner ring and the match began when both robot programs were run simultaneously. The matches were limited to 60 seconds. To win, one robot was required to push the other outside of the outer ring. If any (solid) part a robot went outside the outer ring the match was over and that robot was declared the loser. A tie occurred when both robots were still in the ring after 60 seconds. [1] Our robot was not designed to have sumo wrestling as its strength so we did not actively seek out our opponent during the matches. We implemented the following algorithm: Start: Back wheel on the corner of the inside ring, front wheels extending into the ring and the body at a 45 degree angle to the ring edges. Steps: Robot moves in a straight line until either the touch sensor is pressed or the light sensor detects a black line. IF (touch sensor is pressed) Robot increases speed and continues moving forward in a straight line light sensor does not look for a black line Else if (black line is detected) Back up 3 inches Turn 180 degrees Repeat from first step until match ends 3

4 Because the only objects in the ring were the two robots, only contact with the opposing robot could activate the touch sensor. The speed increased when the touch sensor was activated in an attempt to push the opponent out of the ring. Disabling the light sensor while pushing the opponent prevented the robot from stopping when it passed the edge of the inside ring which would have abandoned the attempt to win the match. Because the touch sensor is quite small, we built a solid wall in front of it for activating the sensor during contact. We had originally intended to utilize the ultrasonic sensor to detect the location of our opponent when in front of our robot. However, the unreliability of this sensor made it an inconsistent method of detecting our opponent. Because it would, at times, detect objects that were not there and at other times fail to detect objects that were there we decided the sensor was not worth using. Minimal testing occurred for this event so it was unknown how it would perform in the competition. During the event the program performed as expected but was not sufficient to beat robots whose design was focused on being sumo champions. Scavenger Hunt The final task in the Robot Olympics was to complete a scavenger hunt. The search area was a rectangle approximately 110 centimetres by 140 centimetres. The robot began the event in the bottom left corner of the rectangle with its back wheel on the corner, front wheels inside the rectangle and the body at approximately a 45 degree angle to the boundaries (see Figure 3). The robot was required to search the area, find two different objects, return to within 10 centimeters of the corner it began in, and stop. The robot was required to emit a sound each time it found an object as well as when it returned to the originating corner and stopped. This task needed to be completed within 5 minutes. [1] We decided to use a search pattern similar to how a room is vacuumed. The following algorithm was implemented: Start: Bottom left corner, back wheel on the junction of the lines and front wheels extending into the search area with the body at a 45 degree angle to the lines Steps: Turn 45 degrees to the left Set objects found to 0 While ( objects found < 2) Move forward until either the light sensor detects a black line or the touch pressed. IF (light sensor finds black line) Back up approximately 2 inches Else if (touch sensor is pressed) Increment objects found variable by 1 Emit noise Back up until light sensor detects black line Check value of objects found sensor is IF ( objects found!= 2) 90 degree turn to the right Follow the black line for 8 inches 4

5 90 degree left turn Else if ( objects found = 2) Turn to the left searching for black line Once black line is found, stop Implement The Long Line Race program to follow the line with the following modification: When turning to the right, time the turn If (time taken > 0.5 seconds) Emit sound End program Else Continue line following The algorithm succeeds when the objects are not placed near each other and when the objects are not placed one behind the other. The distance the robot follows the line after each sweep (in this case 8 inches) needs to be adjusted to suit the size of the objects being searched for. The distance traveled needs be sufficient for the robot to not touch an object a second time once it has been found. The search pattern also relies on the robot making fairly accurate 90 degree turns and driving in straight lines. It was our feeling that using this search pattern made the task of returning to the starting position as easy as possible. It was also the easiest method we could find for ensuring that the two objects located were distinct. We choose to sweep the 110 centimetre distance because of the difficulty having the robot maintain a straight path. The less distance it had to go in a straight line, the less chance it had of veering off that path. It is important to note that the light sensor needed to be moved farther away from the wheels of the robot for this event. Because the surface was uneven, having the light sensor close to the wheels put it too close to the surface, causing it to touch on the raised areas and receive a false light reading. Moving it forward and adding a rigid guard that was marginally lower than the sensor prevented false readings from occurring. During testing the robot completed the required task efficiently. There were some issues with inaccurate turns resulting in the sweeps of the area not being completely perpendicular to the baseline but the majority of the time this was not an issue. With objects close together this would become more of a concern. On competition day the same issues arose but did not prevent the robot from having a successful run. Conclusion and Future Work While we see areas in which the programs could be improved and changes to the robots we feel would make its performance better, the programs as written were successful in allowing the robot to participate competitively. In The Long Line Race the robot had some difficulty with the tighter turns at the beginning, adding valuable seconds to its time. The robot was one of the few in the competition that completed the race in each of its three runs. The fastest time of 1 minute 47 seconds resulted in a second place finish for the Race, a mere 10 seconds behind the winner. The use of two light sensors, one on either side of the black line (approximately three to four inches apart), watching only for the black line and moving away from it when found might see an improvement in the race time. The Scavenger Hunt was a success and warranted the time spent developing this program. With a time of 1 minute 31 seconds, the robot was the first to complete the event but ended up taking second place. Only the first of the two runs was successful. On the second run the robot found the second object, returned to the line, headed for home and then somehow lost the line and ended up turned 180 degrees, disqualifying 5

6 the run (it would never manage to stop at its starting location). The cause of the line loss remains unknown as it was the only time both in testing and competition that this happened. As expected, having the robot drive perpendicular to its reference line was a challenge. Occasionally it would succeed in making accurate turns and driving straight, but not often. It was fortunate that this did not cause enough of an issue to prevent completion of the task. In hindsight, the speed of the motors should have been increased when the robot was following the line back to its starting location to decrease the completion time for the task. Having a reliable ultrasonic sensor would make locating the objects easier and prevent a significant amount of time being spent sweeping the area. An internal mapping mechanism would also help with differentiating between the two objects in the search area and with returning to the starting point. The Sumo Wrestling matches were not successful for the robot. The program operated as expected, but was not enough to be a challenge to robots that were designed with a focus on Sumo Wrestling. A better physical structure for the robot combined with a more robust program would likely put forth a better showing. In one match the robot was knocked onto its side and moved itself out of the ring because the motors for the wheels kept turning. Creating a robust program to perform this task with the limited sensors available for the robots would be an extremely difficult challenge. It is interesting to note that in the majority of the Sumo Wrestling matches the outcome appeared to come down to chance. Robots designed specifically for this task were not guaranteed to beat those that weren t. Given the limited time and resources preparing for the Robot Olympics, the programs performed well. While there are many potential changes to both the physical structure of the robot and the design of the programs that could improve the overall performance, the programs used met the challenge and resulted in a tie for first place in the Robot Olympics. References [1] C. Adsett, A. Brodsky, B. MacKay, and T. Trappenberg, CSCI 1106 Animated Computing Course Notes. Nova Scotia: Dalhousie University,