Metamorphic Robots Seminar Report 2006 ABSTRACT

Transcription

1 ABSTRACT Metamorphic robots with shape changing capabilities provide a powerful and flexible approach to complex tasks in space exploration. Such robots can autonomously and intelligently reconfigure their shape and size to accomplish missions that are difficult or impossible for robots with fixed shapes and configurations. Metamorphic robots are robots that can self-reconfigure their own shape, size and configuration in order to accomplish complex tasks in dynamic and uncertain environments. These systems claim to have many desirable properties including versatility, robustness and low cost. Each module has its own computer, a rich set of sensors, actuators and communication networks. However, the practical application outside of research has yet to be seen. One outstanding issue for such system is the increasing complexity for effectively programming a large distributed system with hundreds or even thousands of nodes in changing configuration. POLYBOT has been developed as its third generation. Dept. of ECE 1 GEC Thrissur

2 CONTENTS 1. Introduction.3 2. Self-reconfiguration through modularity Three promises of n-modular systems Generations Hardware design Structure and actuation Distributed control based on digital hormones Locomotion of robots Configuration-adaptation for environments and tasks Applications Future improvements Conclusion References..26 Dept. of ECE 2 GEC Thrissur

3 1. INTRODUCTION Robots out on the factory floor pretty much know what s coming. Constrained as they are by programming and geometry, their world is just an assembly line. But for robots operating out doors, away from civilization, both mission and geography are unpredictable. Here, robots with the ability to change their shape could adapt to constantly varying terrain. Metamorphic robots are designed so that they can change their external shape without human intervention. One general way to achieve such functionality is to build a robot composed of multiple, identical unit modules. If the modules are designed so that they can be assembled into rigid structures, and so that individual units within such structures can be relocated within and about the structure, then self-reconfiguration is possible. These systems claim to have many desirable properties including versatility, robustness and low cost. Each module has its own computer, a rich set of sensors, actuators and communication networks. However, the practical application outside of research has yet to be seen.one outstanding issue for such systems is the increasing complexity for effectively programming a large distributed system, with hundreds or even thousands of nodes in changing configurations. PolyBot has been developed through as third generation at the Xerox Palo alto Research Center and Conro robots built at the information sciences institute at the University of Southern California are examples for metamorphic robots. Dept. of ECE 3 GEC Thrissur

4 2. SELF-RECONFIGURATION THROUGH MODULARITY Modularity means composed of multiple identical units called modules. The robot is made up of thousands of modules. The systems addressed here are automatically reconfiguring, and for this the hardware systems that tend to be more homogenous than heterogeneous. That is the system may have different types of modules but the ratio of the number of module types to the number of modules is very low. Systems with all of these characteristics are called n-modular where n refers to the number of module types and n is small typically one or two. (e.g. a system with two types of modules is called 2-modular ). The general philosophy is to simplify the design and construction of components while enhancing functionality and versatility through larger numbers of modules. Thus, the low heterogeneity of the system is a design leverage point getting more functionality for a given amount of design.the analog in architecture is the building of a cathedral from many simple bricks in which bricks are of few types.in nature, the analogy is complex organisms like mammals, which have billions of cells, but only hundreds of cell types. 3. THREE PROMISES OF N-MODULAR SYSTEMS 1. Versatility Versatility stems from the many ways in which modules can be connected, much like a child s Lego bricks. It can shape itself to a dog, chair or to a house by reconfiguration. The same set of modules could connect to form a robot with a few long thin arms and a long reach or one with many shorter arms that could lift heavy objects. For a typical system Dept. of ECE 4 GEC Thrissur

5 with hundred of modules, there are usually millions of possible configurations, which can be applied to many diverse tasks. Modular reconfiguration robots with many modules have the ability to form a large variety of shapes to suit different tasks. Figure 2 shows robot in the form of a loop rolling over a flat terrain. Figure 3 shows an earthworm type to slither through obstacles.. Finally Figure 4 shows a spider form to stride over bumpy or hilly terrain. Even though the versatility gives the capability to do a large set of specific tasks it is not necessarily reasonable to use the technology for that task because the tools made specifically for a task are cheaper and more efficient at that specific task than a versatile tool capable of doing many different tasks. For example, an adjustable wrench can be used to tightening a variety of bolts, but a box wrench specifically designed for a particular bolt will work more reliably and cost less. Sometimes versatility is critical. Typically, these are situations in which some information about the environment is not known a prior. Examples of such applications include planetary exploration, undersea mining, search and rescue and other tasks in unstructured unknown environments. The adjustable wrench is a good example. When there are a large variety of bolts to tighten and the performance pf the adjustable wrench is acceptable, an adjustable wrench can be less expensive than an entire set of box wrenches. Dept. of ECE 5 GEC Thrissur

6 FIGURE 1 FIGURE 2 FIGURE 3 VERSATILITY OF MODULAR SELF-RECONFIGURABLE ROBOTS Obviously, turning a bunch of uniform modules into a versatile robot is not child s play. To put together a useful system, solutions must be Dept. of ECE 6 GEC Thrissur

7 found to the complexities of programming a great many coupled, but independent, robotic units. Worse, as more modules are added, many of the programming issues get exponentially harder. These include controlling and coordinating modules to work together effectively and not collide or otherwise interfere with each other. 1. Robustness Another result of being modular and self-reconfigurable is the ability of the system to repair itself. When a system has many identical modules and one fails, any module can replace it. As the number of modules increases, the redundancy also increases. Having redundancy doesnot necessarily increase the robustness of the system. More modules means that there are more modules that cam fail. If a system has millions of modules, it is likely that many of them will not be working properly. There are two basic strategies to increase the robustness to failing modules. The first is to use the redundancy of a system and global feedback to compensate for local errors of individual modules. The classical feedback control view would be that the failed module inserts some disturbance into the system and the global control of the system compensates for the introduced error. The second strategy is sometimes called self-repair. In some instances it maybe appropriate to eject a failed module (detach it) from the system and replace it with a working module from a non-critical position. If a module fails in such a way that the ability to detach itself is also lost, the working modules that are attached directly to the failed module may detach and carry the failed module away. Dept. of ECE 7 GEC Thrissur

8 2. Low Cost One of the general tenets of the modular reconfiguration approaches is that versatility comes from the programming of the devices. Hence rather than developing unique hardware and then programming it for a given robotic task, the program is instead reduced to (re) programming the existing versatile hardware. The broad utility of this method will require the development of programming tools to facilitate and simplify programming. The cost of programming (and reprogramming) systems is often more than the cost of the hardware, thus reducing the value of the flexible nature of the hardware. 4. GENERATIONS Generation 1(G1) which is a simple quick made prototype with hobby RC servos. G1 was built mainly to prototype and evaluate a variety of configurations and gaits. Its modules are of one type only, a hinge, and because they cannot dock and undock automatically, they must be plugged together by hand. It is powered by on-board batteries and controlled by an 8-bit microcontroller, the PICI6F877 from Microchip Technology Inc. The structure was built using laser-cut plastic parts. Up to 32 modules were bolted together and controlled via gait tables with off board computing. G1 is pictured in figure 5. Generation 2 (G2) functionality adds self-reconfiguration capability, additional strength and on-board computing. G2 is made of just two types of cube-shaped modules: segments and nodes. G2 is pictured in figure 6 Dept. of ECE 8 GEC Thrissur

9 Generation 3 (G3) is currently under development. It is much more compact than G2 and adds a brake/ratchet to the main actuation. G3 contains the segments and nodes with other proximity, tactile, and force/torque sensors, plus possibly a low-resolution CMOS camera. These sensors and camera will help the robot with manipulating objects and interacting with the environment. G3 is shown in figure 7. FIGURE 4 FIGURE 5 Dept. of ECE 9 GEC Thrissur

10 FIGURE 6 5. HARDWARE DESIGN Each module has on-board computational power; the on-board battery power; the Ability of modules to automatically dock, attach, and detach themselves and the power of the modules motor. The modules may be either node type or segment type.the segment module has 1 DOF and 2 connection ports. The node module is rigid with no internal DOF and 6 connection ports. Module should fit within 5cm on a side. A segment that has a hinge joint between two hermaphroditic connection plates, and a node, which doesn t move but has six connection plates. Most of the functions depend on the hinged segment, which produces the robots movement, whereas the node s job is to provide branches to other chains of segments. In theory with enough nodes and segments, robot could approximate any shape Dept. of ECE 10 GEC Thrissur

11 DISSECTING THE MODULE The segment module can be divided into three subsystems: 1. Connection plate 2. Sensing, computation and communication 3. Structure and actuation. Connection Plate Each segment has two connection plates. The connection plates serves two purposes, one is to attach two modules physically together. The other is to attach two modules electrically together as both power and communications are passed from module to module. The robot allows two connection plates to mate in 90 degrees. This multi way attachment requires the electrical connectors to be both hermaphroditic as well as 4 times redundant. The connection plate consists of 4 grooved pins along with 4 chamfered holes as shown in fig 3. An SMA actuator rotates a latching plate that catches the grooves in the four pins from a mating connection plate. FIGURE 7 Dept. of ECE 11 GEC Thrissur

12 Each connection plate has 2 photo diodes and 4 LED s that are sequenced to allow the determination of the relative 6 DOF position and orientation of a mating plate. These facilities closed loop docking of two modules and their connection plates. Various sensors are infrared [IR] sensors, accelerometers, potentiometers, force, and touch sensors. These sensors are used to determine the current state of the system and its environment; to obtain the six degree-of-freedom [x, y, z, pitch, roll, and yaw] offset between two docking plates for automatic reconfiguration; to select the right gait for locomotion; and to trigger different behavior modes in response to different terrain conditions. The G2 has two kinds of sensors: 1. Position sensors and 2. Proximity sensors. The first are Hall effect sensors, to determine the angle between the two connection plates. These sensors measure voltage induced by magnetic flux to determine the motor s angle with a resolution of 0.45 degrees. These also serve for commutation and are built into the segment s 30 w brush less DC motors, which can generate 4.5 Newton-meters of torque. The proximity sensors are infrared detectors and emitters mounted on the connection plates. They serve primarily to aid in docking two modules but can also be used to help the robot maneuver in tight spaces. In G3, sensing will include the BLDC Hall effect sensors as well as a joint angle potentiometer, tactile whiskers, tension sensors on the interface pins and accelerometers for orientation and potentially bump. Dept. of ECE 12 GEC Thrissur

13 The new design places four emitter detector pairs on the center of four edges shows the new mechanical design of the plate, the dot denotes an emitter and the circle denotes a detector. The new design has the property that when two centered faces are closer, the intensities received from the corresponding emitters are larger; while in the previous design, the intensities are all diminished and eventually vanished due to large emitter- detector angles. The new design also enables local communication for two connected modules. There are two accelerometers on each PolyBot module. The two accelerometers are mounted on two orthogonal planes, one on x-y plane, one on YZ plane. The x-y plane is fixed to the module base; the Y-Z plane can be rotated about the z-axis. There are four readings, two from each accelerometer. The value of accelerometer reading reflects acceleration along the axis. Each module contains a Motorola PowerPC555 embedded processor with 448k internal flash ROM 1 megabyte of external RAM. This is a relatively powerful processor to have one very module. Each module communicates over a local bus within chains of segments using the [controller area network] CAN bus standard. The six sided nodes will have switching and routing capability to pass messages from segment chain to segment chain. Two CAN buses on each module allows the chaining of multiple module groups to communicate without running into bus address space limitations. 5. STRUCTURE AND ACTUATION The 25-sq. cm connection plate showed on this Poly Bot G2 segment mates with an adjacent module. Infrared sensors align the modules for Dept. of ECE 13 GEC Thrissur

14 docking and a latch made of shape-memory alloy holds them together. Holes and pins add stability to the connection, with power and data transmitted via electrical connectors. under the hood where thay cant be seen are the microprocessor and memory. For physically docking and undocking, every connection plate also houses a latch. At it s heart the latch is a wire made of a shape memory alloy, a nickel-titanium combination that alternates between two shapes when alternated between two temperatures. In this case, resistive heating is used. When current is run through the wire, the latch opens and releases its hold on a neighboring module, stopping the current allows the latch to close by a return spring. An SMA actuator returns a latching plate that catches the grooves in the four pins from a mating connection plate. FIGURE 8 : Segment module of a Conro robot In the Conro system, every module moves every other, with two small hobby servomotors that actuate right angled hinges controlled by an 8-bit micro controller. The modules communicate with their neighbors through an infrared interface. Rather than hermaphroditic connection plates, Conro s modules have three male connectors at one end and one female connector at the other. A system like this will easily form tree-like structures (the same structure as limbed animals) as well as structures with single loops but none with more than one loop no figure-8s. Dept. of ECE 14 GEC Thrissur

15 The segment structure of PolyBot consists of two frame elements, which rotate relative to one another and carry the connection plate components, the actuator and the electronics. This can be rotated up to +90 or 90 degrees. A brush less DC motor with gear reduction sits in the middle of the segment on the axis of rotation and actuates this single DOF. Hobby servos were used in the G1 versions. The standard size servos used deliver maximum torque of 0.7 Nm with torque density up to 11 Nm/kg. While these hobby servos came in a variety of sizes and are easy to interface with both electrically and mechanically they are somewhat under powered and fragile for this application. More torque and robustness were desired for G2. An off the shelf MicroMO gear motor was selected which could deliver 5.6 Nm of torque. This gear motor has a torque density of 19 Nm/kg, and was satisfactory in many respects but weighs 300g and is about 110mm long. It was desired that the G3 modules confine to a 50mm x 50mm x 50mm volume limit, i.e, a 5 cubic cm. It uses a modified Maxon 32mm diameter brush less pancake motor as the source with a 3.75:1 planetary gear stage between the motor and the size 8 100:1 harmonic output stage. This new main drive weighs only 70 grams compared to G2 s 300g bringing the total module weight down from 450 g to about 200 grams. The G3 drive should deliver 1 Nm of torque and the machined aluminum frame has a range of motion of +90 to 90 degrees. In addition an actuator roller ratchet will provide Nm of braking in either direction. 6. LOCOMOTION OF ROBOTS How the robots move is determined by the angle between the connection plates that each modules motor makes. The angles are downloaded from Dept. of ECE 15 GEC Thrissur

16 a gait control table; result in a sequence [top to bottom] that propels the robot. In addition to the physically implemented gaits, several further gaits have been simulated: a 4 armed cartwheel locomotion, exotic gait: carrying an object while rolling, a rolling loop with many feet on the outside rolling/walking, slinky locomotion moving on an x-y grid. Since locomotion is essentially a dual of manipulation, many of the legged gaits were demonstrated to show manipulation of objects. The sinusoid snake like locomotion was demonstrated to work over a variety of objects including crawling in 4 diameter aluminum ducting pipes, up ramps (up to 30 degrees), over chicken wire, climbing 1.75 steps, over loose debris and wooden pallets. In crossing obstacles with a single chain of modules like the sinusoid locomotion, 2 properties were determined to be essential. one is characteristic torque, a unit less quantity indicating the number of modules that can be raised to a cantilevered condition. In order to cross large obstacles, like climbing stairs, the actuators need to supply large torques. For stairs, torque enough to lift about 0.3 m worth of modules would be useful. The other property is compliance, compliance within the modules is useful for the system to conform to the terrain and gain maximal foot contact. For highly geared systems these two properties often conflict. The G1 modules with a characteristic torque of less than 5 do not have enough torque to demonstrate some gaits, but with their proportionally controlled, back drivable servos naturally conform to terrain and duct work. The G2 modules have a characteristic torque of 8 and PID control giving them little compliance. G3 will have a characteristic torque close to 6. The additional sensors on G3 should facilitate some form of active compliance to terrain while an actuated ratchet mechanism will provide large static torque on demand. Dept. of ECE 16 GEC Thrissur

17 Figure illustrates some capabilities of metamorphic robots, where a snake robot just passes through a pipe and is considering what configuration to use in order to manipulate the encountered cubic object. A unique and perhaps the most desirable feature of metamorphic robots is their robustness against damages to individual modules. Modularized self-reconfigurable robots could perform self-repair when failures occur to individual modules by discarding the damaged modules via reconfiguration. FIGURE 9 Dept. of ECE 17 GEC Thrissur

18 7. DISTRIBUTED CONTROL BASED ON DIGITAL HORMONES The main idea of Digital Hormones is that biological hormones are signals that do not require cells to have addresses or identifiers, yet can trigger different cells to perform different actions at different sites. Such signals propagate in a global medium, yet preserve the autonomy of each individual cell. They are different from the pheromones approach because they do not leave residues in the environment. A self-reconfigurable system consists of two basic types of elements: a set of cells and digital hormones. Each cell is an autonomous agent that has certain properties. One major property is that it can secret digital hormones and has receptors to digital hormones. Digital hormones are typed elements that can be released from cells and captured by receptors of neighboring cells. Each type of digital hormone has its own density threshold and diffusion function. Digital hormones are propagated in the space from the higher density space to the lower density space with the ratio specified by the diffusion function. The propagation stops when the density is below a threshold. The receptors have different receptor types and can bind digital hormones with matching types. Receptors can be used up when bound to digital hormones, and can be created or deleted by cell s actions. Thus the number and types of receptors in a cell may vary in a life time. For simplicity, actions of cells are adhesion, migration, secretion, modification, and proliferation. In the context of metamorphic robots, each selfreconfigurable robot is made of robotic cells (r-cells) that can connect and disconnect with each other to form different configurations or organizations. Dept. of ECE 18 GEC Thrissur

19 The assumption is such that the r-cells have the similar actions as those proposed above but can physically connect to one another. All r-cells have the same internal structure as shown in Figure 2, where software and hardware are constructed to simulate the biological receptors and the relevant part of decision-makings process. A local engine with a set of receptors can examine the incoming signals received from its active links, and decides if any local actions should be taken. Such actions include activating local sensors and actuators, modifying local receptors or programs, generating new digital hormones, or terminating digital hormones. Just as a biological cell, a r-cell s decisions and actions depend only on the received hormones, its receptors, and its local information and knowledge. FIGURE 10 The representation of a configuration of r-cells is a graph, where nodes are r-cells and edges are established connections. For example, a single CONRO-like r-cell with four potential connectors can be represented as the graph shown in Figure 3(a) where all four connectors are open. A graph for a snake-like chain of four r-cells can be represented as a graph in Figure 3(b), a 6-legged insect in Figure 3(c), and a system with two separate robots with a remote communication link (dashed line) in Figure 3(d). Dept. of ECE 19 GEC Thrissur

20 In general, an organization of r-cells can be an arbitrary graph with r-cells having different number of connections. FIGURE 11 : Example of r-cell organization The digital hormones can be used to accomplish the communication, collaboration, and synchronization among r-cells. From a computational point of view, a digital hormone is a message propagating in the r-cell network and it has three important properties: (1) a hormone has no destination; (2) a hormone has a lifetime; and (3) a hormone contains codes that can trigger different actions at different receiving r-cells. To illustrate the application of DH-Model in metamorphic robots, consider an example how digital hormones are used in self-reconfiguration. A metamorphic robot with seven r-cells changes from a quadruped (a four legged structure) to a snake. In the figure, a r-cell is represented as a line segment with two ends: a diamond-shaped end (the back link) and a circle-shaped end (this end has three possible links: the front, left and right). The robot must change from a legged configuration (at the top-left of the figure) into a snake (at the bottom of the figure). To do so, this robot must perform the leg-tail assimilating action four times. To assimilate a leg into the tail, the robot first connects its tail to the foot of a leg and then disconnects the leg from the body (shown at the upper part of the figure). Just as in any r-cell organization, each r-cell in the robot determines its role based on its local state information including its own neighboring connections. Using digital hormones, the entire reconfiguration procedure Dept. of ECE 20 GEC Thrissur

21 starts when one (and any one) of the r-cells generates a reconfiguration digital hormone LTS (Legs To Snake). FIGURE 12 This LTS digital hormone is floating to all r-cells, but each r-cell s reaction to this LTS Digital hormone will be different depending on the receiver s role in the current configuration. For this particular digital hormone, no r- cell will react except the foot rcells, which will be triggered to generate a new digital hormone RCT (Requesting to Connect to Tail). Since there are four legs at this point, four RCT digital hormones will be floating in the system. Each RCT carries a unique signature for its sender. No r-cell will react to a RCT digital hormone except the tail r-cell. Seeing a RCT digital hormone, the tail model will do two things: acknowledge the RCT by sending out a new TAR (Tail Accept Request) digital hormone with the signature received in the RCT, and inhibit its receptor for accepting any other RCTs. The new TAR digital hormone will reach all rcells, but only the leg r-cell that initiated the acknowledged RCT will react. It first terminates its generation of RCT, and then generates a new digital hormone ALT (Assimilating a Leg to the Tail) and starts the required reconfiguration. When seeing an ALT digital hormone, the tail r-cell will terminate the TAR digital hormone and start actions to assimilate the leg. After the action is done, the tail r-cell will reactivate its receptor for RCT Dept. of ECE 21 GEC Thrissur

22 digital hormones, and another leg assimilation will be performed. This procedure will be repeated until all legs are assimilated. 8. CONFIGURATION-ADAPTATION FOR ENVIRONMENTS AND TASKS Self-reconfiguration in a dynamic environment is a very challenging problem. In order to change its current shape to fit the environment for missions, a self-reconfigurable robot must sense its surroundings and evaluate its current configuration to see if it is suitable for the mission goals. If not, the robot must search through the space of possible configurations, and find one that is appropriate for the current situation and then change its shape to that configuration. For example, if a legged robot must go through a thin pipe, it must realize that the size of its current shape is too large for the pipe and it must search through the configuration space and determine that it must reconfigure into a snake. Clearly, various coordination tasks must be solved in this process. These include how to represent, research, evaluate, select, and execute configuration changes. Since each configuration is represented as a graph, the space of configurations is represented as a state space of graphs. In this search space, each state is a configuration graph and it can be changed to another state through an operator. An operator is a dock/disconnect action that can either adds a new edge or deletes an existing edge. For example, by connecting the tail to the right-rear foot, a four-legged robot can become a configuration of three legs with a looped tail. Note that module types capture the important Dept. of ECE 22 GEC Thrissur

23 configuration properties of modules and such information can be exploited to evaluate the fitness of configuration to the current environment and mission. Configuration graphs in this search space are evaluated by an evaluation function. This function takes a set of configuration properties and a set of environmental properties and returns a value in a total ordered domain. The configuration properties include the configuration s size, elements types, maneuverability, energy consumption, and functions, while the environmental properties include the nature of the terrain (rugged, smooth, tilted, flat, and so on), and the height of the obstacle, and the size of the passage (e.g., the height of a tunnel or pipe). It is assumed that these properties are sensed from the environment by sensors. Thus, once facing a new situation, a self-reconfigurable robot first discovers its current configuration topology, evaluates the situation with its current configuration using the evaluation function. If the return value is below a threshold, it will search through the state space of configuration by mentally generating (using the operator) and testing (using the evaluation function) new configurations until a satisfying configuration is found. The entire process will be implemented in the Digital Hormone framework to exploit the advantages from both computational and biological systems. Dept. of ECE 23 GEC Thrissur

24 9. APPLICATIONS Planetary explorations Undersea mining Maintenance tasks on ship hulls and oil tanks Inspection and rescue : In such places as pipes, nuclear plant, or tunnels, which are hardly accessible to humans Other tasks in unstructured unknown environments. Dept. of ECE 24 GEC Thrissur

25 10. Future Improvement Currently, all the modules are controlled by a host computer in a central manner to perform previously planned motions. In future, technology will made to allow the modular robot to move or conduct tasks by adapting itself to unknown or dynamically changing environment through the following improvements: Equipping modules with sensors to detect external environments Implementing a distributed control system where each module decides its action in an autonomous way Automatic and more efficient motion planning that is currently hand-coded. Improvement of connection and motion mechanism for more reliable motion. Dept. of ECE 25 GEC Thrissur

26 11. CONCLUSION Metamorphic robots are a revolutionary robotic technology that could greatly enhance human capabilities in space and enable a new range of scientific activities and planetary exploration. In particular, this technology opens the possibility of constructing robots that can change their shape and functionality, thereby affording extreme adaptability to new tasks and changing environmental constraints. These modular systems could be employed by astronauts to construct robotic devices suited to new tasks. At a more advanced stage, this technology could allow for autonomous reconfiguration, and this possibility has also been demonstrated with the concept of Digital Hormones, a communication and control system inspired by biology that provides each module with the information necessary to decide on needed reconfiguration actions. Dept. of ECE 26 GEC Thrissur

27 12. REFERENCES parasol.tamu.edu www2.parc.com custer.me.jhu.edu Dept. of ECE 27 GEC Thrissur