Feb 19, 2007 - A design for a modular centipede robot that has the ability to reconfigure is proposed. The system will be able to autonomously form different ...
Self-Configuring Modular Centipede Robot Don Miner, John Glaros and Tim Oates University of Maryland, Baltimore County February 19, 2007
Abstract A design for a modular centipede robot that has the ability to reconfigure is proposed. The system will be able to autonomously form different numbers of centipedes of different lengths as conditions of the environment change. The system is designed with the goal of being able to traverse and explore rough terrain. Modules consist of four legs and a two-degree freedom torso. Also, a worm gear mechanism on each module is used to form solid connections with others. This system is classified as a swarm robot, opposed to a reconfigurable robot, because each individual module controls itself. Concepts from swarm intelligence, robotics, machine learning, mobile ad-hoc wireless networks, distributed systems and team formation may be applied to this system.
Figure 1: An MCR module.
1. Introduction SwarmBot by iRobot have designs that are only favorable in perfectly flat environments. Our design’s usage of legs is superior to the small wheels and treads seen in previous projects on rough terrain. Only one of these, the SWARM-BOTS project, has a mechanism to physically join the robots[2]. Our design has a stronger method of joining the robots that requires no energy to sustain once joined. Previous work has been done in simulating natural centipede movement [3]. We will use this research when designing a gait for multi-module configurations.
We propose a multi-robot system that is capable of traversing and exploring rough terrain. The central idea to our approach is forming rigid connections between individual legged modules in a linear fashion, to form a robot biologically inspired by centipedes. This design has several advantages, which are discussed later in this paper. This paper is an overview of the Modular Centipede Robot (MCR). We describe the essential individual module design principles and its possible capabilities. We then discuss how a multi-module centipede would function. Other than the design of the robot, several possible research opportunities outside of robotics are possible in this project, which are outlined.
3. Module Design Each module in the MCR system is designed around a few simple essential components. These components include legs, a powered torso and a powered mechanism to join modules together. Final designs have not been decided upon, however, currently each module is planned to have four legs with two degrees of freedom, a torso with two degrees of freedom and a worm gear. Figure 1 depicts the current concept design of an individual module.
2. Previous Work Other projects such as SWARM-BOTS[1] have provided techniques for implementing and controlling robot swarms. However, this project and others have not designed their robots to be able to traverse and explore rough terrain. Robots such as the ones in SWARM-BOTS, the Khepera line of robots and 1
Figure 4: Each module has a tapered screw that can join with a threaded hole in other modules. Figure 2: A module’s leg has two degrees of freedom.
Figure 5: Two modules that are joined together. that has threads on the inside and a male part that is a cone-shaped screw that tapers off at the end. The male part is at the back of the module and the female part is at the front. This is to prevent the screw getting caught in terrain while the module is moving forward. Figure 4 shows the basic design of the screw. When connecting, one module approaches the other and the screw begins to rotate. Once the screw has entered completely, the two modules form a solid connection that requires no power to sustain. Two connected modules are shown in Figure 5.
Figure 3: A module vertically flexing its torso.
3.1. Legs The legs of the module are rather simple and have two degrees of freedom. Each leg may be rotated 360◦ at its joint with the body and then may be rotated around its knee. The basic design of the leg is depicted in Figure 2. The current design uses four legs, one at each corner of the body. Using six legs per module for more individual stability is under consideration. The legs do not need an additional degree of freedom to allow turning because the torso swivels side to side.
3.4. Body The main part of the body is segmented into two parts, joined b1y the torso. The body itself should be as small as possible while including the necessary control boards, power supplies and motors. Being slender will give more room for additional specialized sensors that may be added later.
3.2. Torso Each module has a single torso that has two degrees of freedom near the center of the body. Two servos are required for this design: one for vertical angular movement and one for horizontal angular movement. The joint that moves from side to side is used for turning and the joint that moves up and down is used for climbing and adjusting to different heights in terrain. The torso moving sideways is shown in Figure 1 and the torso moving vertically is shown in Figure 3.
3.5. Communications Each module has to be able to communicate with others in an efficient manner. For inter-module communication we decided to use bluetooth due to the low power consumption. Typical radio frequency wireless devices use a lot more power than bluetooth; often times bluetooth barely affects battery life of a device at all [4]. The only downside is bluetooth’s range is significantly lower than that of RF, but since the modules will be working in close proximity this should not be a problem. An important part of communication is having
3.3. Physical Connections The feature of each module that is required to join with other members of the swarm is the worm gear mechanism. Each module has a female part of the joint 2
each module be able to classify other modules in the following different ways: • Other module is attached to module’s male connector Figure 6: A seven module centipede.
• Other module is attached to module’s female connector • Other module is before the module in the chain • Other module is after of the module in the chain • Other module is not a member of the module’s chain
Figure 7: A seven module centipede moving over a hill. Notice the modules adjusting at the center of their body.
Classifying modules into different categories is important because tasks to be performed in MCR will require coordination. For example, a module may need to cue the module behind it to turn right after it turns right. This classification is hard to do with just wireless communication, even if the the exact position of other modules are known. To make this identification possible, modules must perform a handshake, passing one another’s bluetooth identifications, when joining. The method for this handshake has not been decided upon yet, but it is known that the handshake protocol is required.
cial Intelligence and other fields. These possibilities are outlined below.
5.1. Robotics In building this robotic swarm, we will take strives to make an inexpensive, small, maneuverable, legged robot. This task in itself is will be interesting. Work will have to be done to find an acceptable way for two robots to physically join together. Although we expect the rotating screw approach to work, we may consider other options. Control of the legs is another interesting topic. It is not feasible to have the robot control every leg individually. Also, modules will have to coordinate with each other to produce movement, since there is not one mind controlling all modules. Actual centipedes are the motivation we will use for our first controller. Centipedes move their legs in a wave motion and the position of legs further down the centipede are completely based on where the first legs were placed [5][3].
4. Modular Centipede Once several modules have formed a chain, the robot resembles a centipede, as in Figure 6. Centipedes have some very beneficial properties for walking on rough terrain. They have many legs, making for several points of contact with the ground that helps with stability. Their bodies can conform to the varying hight of the landscape, which is good for preventing the robot from getting stuck in the environment (see Figure 7). Since their bodies are long, the front of the body is able to lift vertically while still being stable and balanced on the ground (see Figure 8). The centipede will be autonomously assembled by the swarm to adapt to the current situation. While MCR is running, there will be an ongoing optimization problem depending on the task. For example, if searching a plot of land, having one long centipede would take a long time to cover everything. On the other hand, having all the modules be independent may yield some getting stuck. MCR will autonomously try to strike a balance between these two extremes.
5.2. Swarm Intelligence There are a few situations where emergent behavior from the simple minded modules will be useful. Can each module’s behavior be modeled as a simple state diagram? If so, walking, choosing when to connect, choosing when to climb could be done with a simple reactive agent protocol.
5.3. Machine Learning
5. Possible Research Topics
Machine learning can be applied to several different aspects of MCR. Reinforcement learning could be used to improve the gait based on changing terrain or based
MCR has the possibility of being the experimental robotics platform of several different areas in Artifi3
tributed of work will have to be used in MCR. Since there is no one mastermind specified in the system, modules will have to make decisions amongst themselves what to do.
6. Conclusion We propose this system, MCR, be designed in detail and then be built, with the goal of being able to traverse rough terrain. With our simple yet agile module design and the ability to conjunction make this system very promising. Along the path to completion, an interdisciplinary effort will have to be made to make the system completely functional and autonomous. Not only do we hope to reach our goal of walking on rough terrain, but we hope that MCR will be a robotics platform that can be used for other artificial intelligence research projects.
Figure 8: A six module centipede climbing a vertical wall. Again, notice the modules adjusting at the center of their body. on different centipede lengths. Learning can be used to have the centipede teach itself how to overcome specific obstacles, such as going over steep inclines by climbing over it. Most learning will have to be done while keeping in mind that each module may be rather simple minded and in most cases most system wide tasks are done through multi-agent emergent behavior.
7. Acknowledgments The graphical designs of the module were done by Matt Ferrero with SolidWorks.
5.4. Mobile Wireless Communication The communications protocol will have to allow for adhoc communication between modules. The existence of a central communications hub, such as a wireless access point, in the environment cannot be assumed. Also, there may not be a direct link between a module and the modules it is attached to. A way must be found for an individual module to figure out which modules it is attached to so that it may communicate important information about global goals of the centipede, such as walking.
References [1] M. Dorigo. Swarm-bot: An experiment in swarm robotics. In P. Arabshahi and A. Martinoli, editors, Proceedings of SIS 2005 – 2005 IEEE Swarm Intelligence Symposium, pages 192–200. IEEE Press, Piscataway, NJ, 2005. [2] R. Groß, M. Bonani, F. Mondada, and M. Dorigo. Autonomous self-assembly in a swarm-bot. In K. Murase, K. Sekiyama, N. Kubota, T. Naniwa, and J. Sitte, editors, Proc. of the 3rd Int. Symp. on Autonomous Minirobots for Research and Edutainment (AMiRE 2005), pages 314–322. Springer, Berlin, Germany, 2006.
5.5. Team Formation If MCR is designed so that different modules contain different sensors and play different roles, team formation strategies will be needed. Modules will have to form centipedes where their specialized skills are distributed optimally. For example, an obstacle detecting module may greatly improve performance. This obstacle detector should be distributed among different centipedes as having two on one may be overkill. Also, a module’s position in a chain is important. A obstacle detecting module would probably be best placed at the front of the centipede. The selection and organization of teams would be done autonomously and different configurations could be learned by MCR.
[3] Viet-Tam Luu. Simulating Centipede Locomotion. University of Waterloo: CS499 Project Report, 1999. [4] E. Ferro and F. Potort. Bluetooth and wi-fi wireless protocols: A survey and a comparison. IEEE Wireless Communications Magazine, 12(1):12–26, February 2005. [5] B. Anders, J. Shultz, and B. Jayne. Axial kinematics and muscle activity during terrestrial locomotion of the centipede Scolopendra heros. The Journal of Experimental Biology, 198:1185–1195, 1995.
5.6. Distributed Algorithms Standard distributed algorithms principles such as election algorithms, multicasting, fault tolerance and dis4