to address this issue, we present a method to carry the Scouts to and from their main ... Arkin and Balch list some of the reasons for forming a team rather than ...
A Method for Transporting a Team of Miniature Robots Esra Kadioglu, Nikolaos Papanikolopoulos Center for Distributed Robotics, Department of Computer Science and Engineering University of Minnesota, Minneapolis, U.S.A. {kadioglu, npapas}@cs.umn.edu
Abstract— The Scouts, developed at the University of Minnesota, are miniature robots designed mainly for reconnaissance and surveillance tasks. The Scout’s small size and multiple mobility and sensing modes allow it to efficiently navigate and carry out certain missions in indoor environments and on relatively smooth surfaces. However, like many other small-sized robots, its miniature size and limited battery life become a bottleneck when it comes to cover long distances, particularly on rough, outdoor surfaces. In order to address this issue, we present a method to carry the Scouts to and from their main mission locations by adding a larger robot, a Pioneer 2-AT, to the team. The Pioneer carries a team of Scouts in a color marked box that it holds with its grippers. The Scouts jump out of the box, carry out their mission, and autonomously find the box and jump back in.
I. INTRODUCTION The design phase of a robot usually begins with a target domain and a task in mind, and the decisions are made based on that specific objective to be achieved. Once the goal is achieved, the general tendency is towards widening the borders of the initial domain. With this change of objective, in most cases, the robot needs to be changed, too; or supported by other robots. Arkin and Balch list some of the reasons for forming a team rather than building an “all-purpose” robot in [3] as : -Distributed Action, -Inherent Parallelism, -Divide and Conquer, -Simple is better. The first two provide better spatial and temporal handling of the task, respectively. The divide and conquer approach, though problem dependent, may speed up task completion for a homogeneous team. For a heterogeneous team, it will allow each robot to carry out the task that it is most capable of. Finally, the last one suggests that the robots in the team are likely to be already available and/or simpler than a comprehensive single robot. However, as there are advantages, there are challenges, too, associated with robotic teams. Some of these issues are communication problems, appropriate task assignment, and cooperation and control strategies to be used. In this paper, we present a method to carry a small team of Scouts to and from their main mission points.
The marsupial team consists of a Pioneer 2-AT and four Scouts. The P2-AT carries the Scouts in a box it holds with its grippers. The Scouts, once their mission is complete, autonomously find the box using their cameras and the color markers on the box, dock, and jump in. The P2AT then carries them to a specified location. Unlike previous papers on the Scout platform, the emphasis of this work is on the coordination of a marsupial team. In particular, we focus on the development of methods that promote effective deployment and retrieval of individual team members. The main motivation behind this implementation is to enhance the functionality of the Scouts. The small size of the Scouts makes them ideal for reconnaissance and surveillance tasks where the robot should be able to hide easily, and search and rescue operations where the robot should be able to fit and move through small openings. However, these very features also make it almost impossible for the Scout to move on rough surfaces for long distances to its main mission location. With the method presented in this paper, we try to address this issue. II. RELATED WORK Research on marsupial robotic teams is relatively new. In a work by Murphy et al., the term “marsupial” is defined as “a collection of mobile robots, where one or more robots are at least temporarily physically dependent on another for directives, transport, power, communication, etc.” [9]. In the same work, “mother” agents are referred to as dispensing agents and the “baby” agents are called as the passenger agents. In [8] and [9], a modified Power Wheels jeep (Silver Bullet) carries a shape-shifting, Inuktun VGTV MicroRover (Bujold). Since Bujold is carried in a compartment inside the Silver Bullet, which it entirely fits, only one passenger agent can be carried at a time whereas our approach can accommodate several robots. Bujold is equipped with a camera but has no on-board power and processing capability. Therefore, it uses Silver Bullet’s resources via a tether. The vision-based docking of Bujold, which uses a color-segmentation based approach, is described in [7]. Anderson et al. present another two-robot marsupial team in [2]. A K2A Cybermotion, Inc. robot (MACS)
Fig. 2.
Fig. 1.
A Ranger with a Scout launcher.
carries RACS, an R3 robot by IS Robotics. Both robots have on-board power and processing capabilities. They are used to perform floor characterization in radiologically contaminated indoor areas. RACS, which is initially on a platform mounted to MACS, is deployed once the robots reach a designated area. Two robots survey the area separately and upon task completion RACS returns to the deployment point, using dead reckoning and sensor readings from a homing beacon. It then gets back on the platform. No visual sensors are involved in the docking. In [12], a modified Pioneer AT (Radbot) and a custom, spherical shaped small robot (Subot) form a team. The paper focuses on vision-based docking of the robots, without giving details on the transportation and deployment of the Subot. Sukhatme et al. describe a marsupial case study in [13] where an aerial vehicle deploys a small radio-controlled car. The AVATAR (Autonomous Vehicle Aerial Tracking and Reconnaissance) robot helicopter, equipped with a downward looking color CCD camera, is used to detect and track an intruder on the ground. It then lands to the point where the intruder is last seen, deploys the car, and the car locates the intruder. The AVATAR then lands and retrieves the car. In this experiment, all robot control was by human teleoperation. Nevertheless, it presents a potential application for marsupial teams. An earlier method of carrying the Scouts to the mission site was by a launcher placed on a modified Ranger (See Fig. 1). The Ranger is based on the ATRV-Jr platform from the RWI division of IS Robotics [5]. The launcher fits 10 Scouts and is capable of launching them up to a range of 30 m. Each Scout is placed into a “sabo”, i.e. a protective shell, to protect it from the effects of the impact.
The Scout shown next to a mouse.
This method may not be efficient for a number of reasons: Although the launcher’s angle and propulsion force can be selected, the Scout will bounce and roll after landing. This might make it difficult to predict the location of the Scout. Furthermore, depending on the destination distance and surface, the sabo may not be able to protect the Scout or it might fail to open and release the Scout. Once the Scouts are launched, it is not possible for the Ranger to pick them up again. The team and the method we present in this paper vary from the relevant work in the following aspects: The Pioneer 2-AT carries multiple Scouts at the same time and in an external container, rather than carrying a single robot in an internal compartment. There is no tether between the P2-AT and the Scouts since all have on-board power and communicate through a wireless medium. Although the Scout is dependent on the P2AT for image processing computations, the robots can function autonomously. Docking of the Scouts to the box (they can approach to any of the three sides of the box) and jumping in is relatively easier than docking a robot into a compartment inside another robot. III. ROBOTIC TEAM Our marsupial robotic team consists of two kinds of robots: A small team of Scout robots and a Pioneer 2-AT. The Scout, developed at the University of Minnesota, is a two-wheeled, cylindrical robot that is 40 mm in diameter and 110 mm in length (see Fig. 2). It has two main modes of locomotion: The wheels on its sides and the spring foot. It can use its spring foot to jump over obstacles with a maximum jump height of 0.25 m [5]. Using its wheels, the Scout can roll along relatively smooth surfaces at a speed of .31 m/s [4]. The general purpose Scout weighs about 200 g and contains a variety of sensors, such as video cameras, accelerometers, tiltometers, and wheel encoders. The Scout that is used in our experiments was equipped with a color CMOS camera with a 5 mm / 65◦ lens angle.
Fig. 3.
The Pioneer 2-AT holding the box.
It pauses after each small rotation and checks if it sees the color marker on the box. With this “look” and then “move” sequence, the Scout operates in an open-loop control [6]. If the color mark is spotted, it turns left/right or goes straight depending on the location of the color marker in the image frame. Once the color marker is at the center of the image frame, the Scout moves straight towards the box until it hits it and stops. It then jumps into the box. The Pioneer again monitors success/failure using its camera. If the color marker is not visible to the Scout in the current image frame, it keeps rotating clockwise and checking for the color marker. For the sake of simplicity, we assumed that the box is not blocked by an object and it is detectable by the Scout once it rotates < 360◦ . A. Software
A specialized Scout may come with the following options: Actuating wheels, grappling hook, and infrared emitters. For more detailed information on these capabilities, one may see [4]. The Pioneer 2-AT was equipped with an on-board, Pentium 133 Mhz computer running Linux, frame grabber, and grippers. The grippers can rise 9 cm and lift objects weighing up to 2.5 kg [1]. We used a color CMOS camera, identical to the one on the Scout, to monitor the box held within the grippers and to detect whether the Scout successfully jumped in and out of the box. Although it is possible to run the image processing code on the P2-AT, we run it on a separate Linux PC. The “carrying box” is a small, open cardboard box that can fit four Scouts. Its dimensions are 11 cm x 8.5 cm x 2.5 cm. There is a pink, rectangular shaped paper mark on three sides of it to be used by the Scout to find the box. It also has a cardboard “handle” attached to it for an easy grip by the P2-AT. The handle’s dimensions are 4.5 cm x 3 cm x 3 cm (see Fig. 3). IV. SCOUT TRANSPORTATION SCENARIO The scenario for carrying a small team of Scout robots in a box held by a Pioneer 2-AT is as follows: The Scout team and the P2-AT are currently at the starting location. We place the Scouts into the box and the box into the grippers of the P2-AT. For the time being, we teleoperate the Pioneer. The Scouts, too, are teleoperated for jumping out and moving away from the box since our main focus is on making them find the box and jump in. The Pioneer moves to a destination point and puts the box to the floor. When a Scout receives the “jump” command, it jumps out of the box and moves away from it. Using the camera that monitors the box, the Pioneer compares the images before and after sending the jump command to make sure that the Scout successfully jumped out of the box. Once out of and away from the box, the Scout starts slowly rotating around itself in a clockwise fashion.
Due to its limited computing power, it is not possible to run the image processing code on the Scout. Therefore, an 800 MHz, 256 MB Pentium III PC running Linux is used to process the images. The input from the Scout camera is transmitted through a wireless connection to the PC. The color markers on the box are tracked as “blobs.” In case of multiple blobs in the frame, the one with the largest area is taken into account. Leftmost, rightmost, and the center x coordinates of the colored region are computed and compared with the center of the frame. The pseudocode is as follows: if (blob.leftX > image.centerX) turn right elseif (blob.rightX < image.centerX) turn left elseif (blob.centerX < image.centerX + and (blob.centerX > image.centerX - 10) go straight.
10)
Color thresholding and computation of blob statistics are done using the ActivMedia Color Tracking System (ACTS) [11]. It allows users to select regions of color in an image by clicking on the region and selecting a set of points or a rectangular area (see Fig. 4 for the ACTS interface). Monitoring of the deployment and retrieval of the Scouts is carried out by a camera mounted on the Pioneer, viewing the inside of the box. The camera takes an image before and after each jump command. In case of deployment, if the number of pixels with dark colors is smaller in the current image then the Scout successfully jumped out of the box. For retrieval, a larger number of dark pixels in the current image means a successful jumpin. Therefore, an image differencing operation is sufficient for monitoring. In case of a deployment failure, the Scout is sent the jump command again. On the other hand, if the Scout fails
Fig. 4.
ACTS interface.
to make it into the box after a jump, it has to re-track the color blob since we do not know the orientation of the Scout - it may be facing away from the box. V. EXPERIMENTS We run a number of experiments indoors, under laboratory lighting conditions. Some experimental results are shown in Fig. 5. In the experiments, we observed 4 main issues: Color tracking, communication, wheel slippage, and balance of the box. Color tracking is notoriously dependent on the lighting conditions. Although we performed our tests indoors, variations of illumination within the lab and the distance of the Scout to the box affected the success rate. Since the camera on the Scout is very close to the ground, once it is near the marker, it receives less light and the color seems darker than it is. The ACTS allowed us to select the target color region before the experiment using the view from the Scout camera. With ACTS, we can specify a number of target regions. By using this feature, we can specify the color to be tracked by holding the Scout at various distances to the box and therefore enhance the marker detection capability of the robot. This is more practical compared to specifying a color range which is highly undependable. Furthermore, this method allows us to have a number of different color markers on the box and
select the one less likely to be found in the environment that the Scout would operate. We have a wireless RF communication with the Scout. The commands are sent digitally and the video is received through an analog transmission. Location of the antenna, communication range, and the battery power level of the Scout affected the success of the experiments. The most repeatedly occurred problem was when the Scout failed to transmit the video. In such a case, the robot could not detect any color blob in the image, and assumed it was not facing the box. As a result, it kept rotating and therefore missed the box. Another problem was with the wheel slippage. If the Scout detects the marker and is directly facing the edge of the box, it executes the go straight command and moves until it hits the box. However, if it is relatively away from the box at this point, the wheel slippage may cause it to miss the box and move until either it hits something else or the command times out. In order to solve this problem, we divided this step into 3 sub-steps: When the Scout decides to go straight, it goes for a small distance, stops and checks if it is still directly facing the box. If not, it turns to left/right to reposition itself. Otherwise, it repeats the step one more time and then goes straight until it reaches the box. The last issue was related to the design of the box. When the P2-AT lifts the box by holding it from the handle, the other end of the box tilts down and slightly touches the floor. Increasing the gripper pressure alleviated the problem to a degree. Although this did not create a problem indoors (on carpeted and concrete floor), it is likely to be a problem on rough, outdoor surfaces. We also encountered a few cases in which the Scout jumped into the box but then bounced out of it. In order to avoid this, we placed a thin layer of foam to the bottom of the box. Increasing the edge height of the box might not be a feasible solution considering that the Scouts will jump into the box at the end of their mission, which implies that they will have a low battery level. In those cases when they bounced out after a jump, they had relatively high battery power levels. In experimental runs without any communication problems, the Scout successfully found the box, docked, and jumped in (See Fig. 5). VI. FUTURE WORK In this paper, we have presented a marsupial robotic team consisting of a Pioneer 2-AT and 4 Scouts. The P2AT carries the Scouts to their mission point in a box that it holds with its grippers, and brings them back to the base after the completion of the task. The issues that hindered the success of the experiments are listed in Section V. Nevertheless, we have obtained promising results and our future work includes the following:
The carrying box is a simple, inexpensive, open cardboard box. It can fit four Scouts but considering that the P2-AT can carry up to 2.5 kg and each Scout weighs about 200 g, we can design a box that can carry more than four Scouts. Robots can be inserted into the box by hand but since they have to jump out at the destination, placement should allow for proper operation of the spring foot. They cannot be placed on top of each other and placing them side by side may not be practical in terms of the shape of the box. There are issues related to the handle length, proper height of the sides of the box, ground clearance of the box when lifted, and balancing it with the P2-AT. We consider adding an omnidirectional wheel to the front of the box that will address balancing and ground clearance issues. We also consider building a box with a ramp and drive the Scouts in, rather than making them jump. Since the Scouts have a limited computational capability, color thresholding and blob statistic calculations are done on a separate PC. We plan on using the P2-AT to perform these calculations for the Scout. Currently, the Pioneer camera is used only to monitor deployment and retrieval of the Scouts. We plan on using it to track the Scouts during their mission and/or during the phase when they search for the box [10]. The P2-AT’s camera will be considerably higher than the Scout cameras and hence will have a much wider view. It can inform the Scouts about possible obstacles in the area and help them achieve a more intelligent motion to and away from the box. In this way, the P2-AT will be a more active member of the team and contribute to the autonomy and task completion performance of the Scouts. Another potential point of improvement is placing color markers on the P2-AT that will lead the Scout to the box. In the current version, the Scout cannot find the box if it is in an area behind the P2-AT. VII. CONCLUSIONS
Fig. 5. A series of images showing the Scout docking and jumping into the box. In (a) the Scout begins facing opposite to the box, (b) the Scout starts turning clockwise, searching for the box, (c) box is on the right side of the Scout, (d) the Scout turns right but now the marker is on the left, (e) the Scout repositions itself by turning left, (f) the Scout approaches the box, (g) the Scout hits the box and stops, (h) the Scout jumps into the box, and (i) the Scout is in the box.
A method to carry a team of miniature robots, the Scouts, is presented. This is accomplished by the addition of a Pioneer 2-AT robot carrying an inexpensive, simple box. The P2-AT carries the box containing a team of Scouts to the mission point, the Scouts jump out of the box, move away, and then autonomously find the box and jump back in. We assumed that the box is visible by the Scouts and not blocked by an obstacle. The P2-AT, which is teleoperated in the experiments, monitors deployment and retrieval of the Scouts and then carries the box back to a specified location. The Scouts have a very low ground clearance and their limited battery life makes it difficult, if not impossible, to cover long distances. By forming a marsupial robotic team (Scouts / P2-AT), it is possible to transport the Scouts to their destinations without wasting valuable battery life. It also allows less human involvement, increases the team
capabilities, and widens the domain of operation for the Scouts. VIII. ACKNOWLEDGMENTS This material is based upon work supported by the National Science Foundation through award #EIA-0224363, the Microsoft Corporation, and the Defense Advanced Research Projects Agency, Microsystems Technology Office (Distributed Robotics), ARPA Order No. G155, Program Code No. 8H20, issued by DARPA/CMD under Contract #MDA972-98-C-0008. IX. REFERENCES [1] ActivMedia, Inc., Peterborough, NH. Pioneer Operation Manual v2, 1998. [2] M. Anderson, M. McKay, and B. Richardson. Multirobot automated indoor floor chracterization team. Proc. of the IEEE Int’l Conference on Robotics and Automation, April 1996. [3] R. Arkin and T. Balch. Cooperative multiagent robotic systems. In D. Kortenkamp, R. P. Bonasso, and R. Murphy, editors, Artificial Intelligence and Mobile Robots. MIT/AAAI Press, 1998. [4] A. Drenner, I. Burt, T. Dahlin, B. Kratochvil, C. McMillen, B. Nelson, N. Papanikolopoulos, P. E. Rybski, K. Stubbs, D. Waletzko, and K. B. Yesin. Mobility enhancements to the scout robot platform. Proc. of the IEEE Int’l Conf. on Robotics and Automation, Washington DC, USA, May 2002. [5] D. F. Hougen, S. Benjaafar, J. C. Bonney, J. R. Budenske, M. Dvorak, M. Gini, D. G. Krantz, P. Y. Li, F. Malver, B. Nelson, N. Papanikolopoulos, P. E. Rybski, S. A. Stoeter, R. Voyles, and K. B. Yesin. A miniature robotic system for reconnaissance and surveillance. Proc. of the IEEE Int’l Conf. on Robotics and Automation, pages 501–507, San Francisco, CA, U.S.A., Apr. 2000.
[6] S. Hutchinson, G. D. Hager, and P. I. Corke. A tutorial on visual servo control. IEEE Transactions on Robotics and Automation, 12(5):651–670, 1996. [7] B. W. Minten. Low-order-complexity vision based docking. IEEE Trans. on Robotics and Automation, 17(6):922–930, 2001. [8] R. R. Murphy. Marsupial and shape-shifting robots for urban search and rescue. IEEE Intelligent Systems, 15(2):14–19, 2000. [9] R. R. Murphy, M. Ausmus, M. Bugajska, T. Ellis, T. Johnson, N. Kelley, J. Kiefer, and L. Pollock. Marsupial-like mobile robot societies. In O. Etzioni, J. P. M¨uller, and J. M. Bradshaw, editors, Proc. of the Third International Conference on Autonomous Agents (Agents’99), pages 364–365, Seattle, WA, USA, 1999. ACM Press. [10] D. P. Perrin, E. Kadioglu, S. A. Stoeter, and N. Papanikolopoulos. Localization of miniature mobile robots using constant curvature dynamic contours. Proc. of the IEEE Int’l Conference on Robotics and Automation, May 2002. [11] P. E. Rybski. ACTS: Activmedia Color Tracking System Manual. ActivMedia Robotics, LCC, 44 Concord Street, Peterborough, NH 03458, USA, 2002. [12] J. Spofford, J. Blitch, W. Klarquist, and R. Murphy. Vision-guided heterogeneous mobile robot docking. SPIE Conference on Sensor Fusion and Decentralized Control in Robotic Systems II, 3839:112–121, September 1999. [13] G. Sukhatme, J. Montgomery, and R. Vaughan. Experiments with cooperative aerial-ground robots. In T. Balch and L. Parker, editors, Robot Teams: From Diversity to Polymorphism. AK Peters, 2001.
