Literature Review PDF

Preview

Summary

Lit Review Final ...

Description

Unit to Unit Inductive Coupling in Swarm Robotics Sean Flanagan Chair: TBA University of Massachusetts Amherst

1

Objective Swarm robotics is one of the most undeveloped topics in computer science and robotics. Current research is wide ranging with some developing novel approaches to the optimization of navigation systems, obstacle detection systems, avoidance systems, and communication systems, while others focus on specific applications of swarm robotic systems, the design of swarm robotic systems, or methods for power management. Power management is currently a major issue facing swarm robotic systems as the operation time of a swarm robotic system is highly dependent on battery capacity, a factor limited by design constraints. Limiting the time that a swarm robotic system can be operational is a major issue; for swarm robotic systems to be effective, they must have a longer operational lifetime. In applications where an operation could take days or even longer, battery capacity and energy storage is crucial. This literature review will investigate the current methods used to achieve ideal swarm robotic behavior, practical systems and models that have been developed, their applications, the issue of power management in swarm robotic systems, and a possible solution: inductive coupling. Swarm Intelligence and Biomimicry Swarm intelligence is defined as “the collective behavior of decentralized, self-organized systems, natural or artificial.” [1] Where collective behavior is the accomplishment of a goal by a group of individuals without a central control structure or unit. Instead of a central control structure, collective behavior arises from the interaction between units and their environment. For robotic systems, these units are usually small and execute simple behaviors. The size of the population and the frequency of interactions between units then leads to the collective behavior that is desired. Fundamentally, the implementation of collective behavior in robotics is a form of biomimicry. Collective behavior is a phenomenon commonly seen in nature, as many natural species have evolved to rely on interactions between members of the population to achieve basic 2

goals such as collecting food, building shelter, or defending against predators. Some examples of animals that exhibit collective behavior are birds flocking to protect from predators, bacteria multiplying to collect food, and fish schooling to confuse predator. However, for swarm robotic applications, a more relevant biological example is an ant colony. When ants leave their colony to search for food (a collective goal), they leave pheromone trails wherever they travel (simple behavior). These trails that differ based on the environment they are searching and whether they were successful in their search for food. This is an example of the simple behavior mimicked by a swarm robotic system. When an ant moves near another ant, they interact using their antennae to “smell” molecules known as cuticular hydrocarbons on the outer shell of each other. This interaction is an example for local communication found in swarm robotic systems. Furthermore, these molecules are also affected by their success in foraging and if an ant has found food any ants that interact with it will follow the pheromone path of the successful ant. More interactions between members of the population will lead to more ants following the path creating a much stronger pheromone trail and soon a swarming of individuals exhibiting the collective behavior: finding food for the colony. There are two different types of swarm robotic systems, homogenous and heterogenous. Homogenous systems consist of units similar in structure and capabilities. These systems mimic the previously mentioned ant colony, where each unit (ant) is clearly very similar in structure and capabilities. These units relay information to each other to make an efficient decision. Heterogenous systems are the opposite; they consist of robots with different functions within the system. An example of a heterogenous system is presented in the paper, “Disaster swarm robot development: On going project,” [3] where 3 unique robots carry different life-supporting materials (water, vitamins, and oxygen) for disaster rescue applications. Heterogenous systems

3

have some localized controls, but they still exhibit collective behavior in large populations. Because swarm robots are based on biological phenomena like swarm intelligence and collective behavior, they are more capable of adapting to complex challenges that would arise in real-life situations. The level of complexity of the situation defines the robustness of the swarm robotic system. Flexibility is defined in terms of the uniform design of units. Because units can be very similar, they are more flexible in terms of substitution or removal from the system as cost is likely to be less of an issue compared to a singular unit which would not be as expendable. Also, in many applications, it is easier to solve a certain problem with a large population of units compare to small scale robotic system. An example of this is evident in the paper, “The development of hybrid methods in simple swarm robots for gas leak localization,” [8] where a swarm robotic system was used to identify the location of an indoor gas leak. This same detection would have taken more time if a single robot system was implemented. Additionally, swarm robots can achieve some complicated tasks which are impossible for a single robot with local communication and cooperation. These include searching extremely large environments, traversing complex and unstructured environments, or even combining to form complex structures. Scalability is defined as the ability for the application of a system to be “scaled” up, with a higher unit count to achieve larger-scale tasks. Path Optimization To create a functioning swarm robotic system, an efficient system for navigation must be implemented. Navigation systems within swarm robotics consist of three different parts: obstacle detection, obstacle avoidance, and unit avoidance. Obstacle detection and avoidance refer to the ability of the unit to traverse its environment while avoiding any obstacles and optimizing its path based on environmental conditions. Unit avoidance refers to the ability for a single unit to 4

avoid collisions with other units in the system. The optimization of these three aspects is what encompasses much of the research done in swarm robotics today. The research conducted by Peng, Zhang, and Huang exemplifies this. In their paper [9], they present a gene regulatory network for “expected pattern adaptation” and obstacle avoidance strategies for swarm robotic systems. The results of the simulations presented in this paper are very similar to the findings of Bao and Zalika’s paper [2] where multiple algorithms are implemented in simulations combining a neural network and fuzzy logic controller to create safe paths for a swarm robotic system to follow. Additionally, the results of Vicmudo’s research paper, also accomplish this goal. In this case, a genetic algorithm was implemented for underwater applications again to optimize the paths taken by swarm robotic units. However, similar to the aforementioned papers, the results of Vicmudo’s research are simulation results that suggest successful navigation of an environment. Although each of these papers seek to accomplish the same goal (successful navigation models), they all employ different methods. These can be roughly classified into 2 categories: artificial intelligence and static models. Artificial intelligence broadly is the attempt to mimic the human brain in a computer model. A neural network as shown in figure 1 is an example of a model containing layers of nodes (or neurons) that use make up an artificial intelligence.

5

Figure 1| A Neural Network: ResearchGate A neural network is “trained” by constant input and resizing of the model and that results in an output that attempts to “learn” from the past outputs to yield the most accurate result. An example of the second model type is a fuzzy logic controller (FLC) like the one used in Bao’s paper, “Obstacle Avoidance for Swarm Robot Based on Self-Organizing Migrating Algorithm.” A fuzzy logic controller, as seen below, consists of a static ruleset to which the input is compared.

Figure 2| Fuzzy Logic Controller: tutorialspoint.com This differs from artificial intelligence as an optimized output is reached and given without any “learning” algorithms. However, these two model types can be used together. In two papers written by researchers at Sriwijaya University [7][8], a combination of a fuzzy logic controller

6

and neural network were used together in a swarm robotic system navigation. The neural network was used to optimize the path and behavior of the units within the system while the Fuzzy Logic Controller was used to avoid obstacles in the and maintain distance between the other units themselves within the testing environment. Practical Systems Although most of the research conducted in the field of swarm robotics consists of simulation and mostly idealized results, some research has been done regarding design and the actual implementation of navigation algorithms into basic swarm robotic systems. In the papers mentioned from Sriwijaya University, a small-scale swarm robotic system was implemented. The units of this system consisted of an Arduino microcontroller (see figure 3) acting as its navigational system running the detection and avoidance algorithms. Additionally, an X-BEE module was used for the 2-way communications between each unit to achieve a collective behavior. The goal of this system was the determination of a gas leak location in an indoor environment. In this paper, the models implemented on the Arduino microcontroller were extremely effective and approached the simulation accuracy obtained in the papers by Peng, Bao, and Vicmudo.

7

Figure 3| Arduino Micro: pololu.com Another example of a practical system comes from the research done at the University of Bridgeport by Patil and Abukhalil. In their paper, “UB robot swarm — Design, implementation, and power management,” they outline the implementation of again a small-scale swarm robotic system with similar components. The UB robotic swarm units also used Arduino microcontrollers for the navigation logic and X-BEE modules for unit communications and interactions. The UB robotic swarm was also successful in terms of navigation in a similar fashion to Peng’s work. When considering the results of these works, it must be noted that there are some key differences between the implementation of a simulation and designing practical systems. Simulations often use ideal parameters that would not be achievable by any systems in the real world. These idealized parameters include unit size, unit weight, population size, computing power, and most importantly total power consumption. In simulations similar to the papers mentioned previously, ideal systems with a small unit size and large population can be implemented. For example, Peng’s paper, “Pattern formation in constrained environments: A swarm robot target trapping method,” states that it assumes: 8

The robots are homogeneous; Only center of mass motion is considered for the robots, that means each robot is seen as a moving point. The sensing and communication range of a robot is limited. However, the information of pattern, targets, obstacle and constraints can be shared through networked communication.

This differs greatly from the implemented systems the robots would have some slight variation, the mass of each unit could not be treated as a point, and the sensing range varies based on the quality of sensor used. Additionally, the cost of materials, mechanical design considerations, and battery size greatly affected researchers’ ability to implement small units and large populations in a real system. Finally, simulations can be performed on computers with a very large computing power compared to that of an Arduino microcontroller. The Vicmudo paper states that simulations were run with an Itel Atom 2GB memory chip, quite a difference from the 512 bytes of memory contained on the average Arduino. All of these idealizations also play into the most import consideration when implementing a swarm robotic system, power management. Power Management The size, weight, and computing power of each unit will determine the amount of power needed to drive it. The size of the impacts the form factor of the battery that can be used and ideally, for a small unit, a small battery would have to be used. Weight is also a consideration. For underwater applications like that of Vicmudo’s paper, the weight of the battery must be taken into account when designing a possibly buoyant unit. Finally, the computational power must be taken into account. The more complicated algorithms implemented, the more power the control board is going to require. The paper from the University of Bridgeport addresses these considerations in detail. First, all of the components used in the design of each unit is accounted for. These include 2 ultrasonic sensors, an infrared sensor, temperature and humidity sensors, 4 servo motors, 1 Arduino microcontroller, 2 encoders, and 1 motor controller. Based on the power

9

requirements for the components, the researchers calculated an average use time of 2.69 to 3 hours. The system was then tested with full load and motion, no load with motion, and powersaving mode with motion to determine the accuracy of these calculations. The power-saving mode extended the operation time of the system by an hour on average whereas the other modes were either average or less than average in terms of operation time. These results are problematic. Successful swarm robotic systems need to be able to have a long operational lifetime. In applications like that stated in the paper, “Disaster swarm robot development: On going project,” to search a disaster zone effectively, swarm robotic units would have to have a much higher operational lifetime, perhaps approaching days instead of hours. One attempt to address the problem of power management, researchers at the University of Manchester in England developed a swarm robotic system that implemented mobile inductive charging pads with 12 charging cells too charge units “on-the-fly” and extend the operational lifetime. The researchers then tested to see the effects of the charging plates on the “perpetual operation” of the system. They determined that they could keep units operational in a 12 hour time frame without the need for battery changes, complete recharging, or adverse effects on the capability of the system. However, system is still not ideal. Inductive Coupling The method used to develop the plates described in the University of Manchester paper is known as inductive coupling. Inductive coupling is a form of wireless charging that involves the close proximity or contact between two thin coils of wire, where one is connected to a power source and one is not. This proximity creates a magnetic field on the second coil, inducing an electric current. A coupling factor is calculated based on the distance, size, and position of the second coil relative to the first coil. A higher coupling factor results in a better exchange of energy 10

between the two coils. Based on the calculated coupling factor, there are two different types of inductive coupling: a loosely coupled or tightly coupled system. In a loosely coupled system, the receiving coil does not receive all of the energy produced by the transmitting coil because the couple does not require all the power being output by the transmitting coil, or the design constrains the size of the receiving coil. Tightly coupled systems consist a receiving coil that has the same dimensions as the transmitting coil. In the wireless charging pads used in the University of Manchester paper as well as many wireless phone chargers today, overlapping transmitting coils are used. This is because it is extremely hard to achieve perfect alignment between one transmitting coil and one receiving coil in a tightly coupled system. Overlapping multiple coils results in a higher coupling factor and in turn a better energy transfer between the coils. Applications The applications for swarm robotic systems are wide-ranging. Because of the scalability of systems, swarm robotics are ideal for large environments with non-uniform obstacles, and environments where human employment is impossible or dangerous. Some of the research papers reviewed presented very specific applications for the swarm robotic systems presented while others remained very general. The two papers from Sriwijaya University, highlight the use of swarm robotic systems in indoor gas leak detection, stating, “For example, gas leak detector, which would be very dangerous if the people who do it directly.” Additionally, Kuswadi’s paper, “Disaster swarm robot development: On going project,” highlights the usefulness of swarm robotics in disaster situations. Because post-disaster zones are very unpredictable and unsettled, it would be extremely useful for small, cheap robots that could cover a large area to be launched for search and rescue missions or aid needs. The paper goes further into the potential applications in the authors home country of Indonesia where there is a “high risk of disasters, such as flood, 11

volcano eruption, landslide, forest fires, earthquake and tsunami.” Some of the more general applications include underwater communication as highlighted in Vicmudo’s work [6] or simply for a very general swarm robotic application as in Peng’s paper [9]

Conclusion After a review of the current swarm robotic models and systems in addition to the design considerations of currently implemented systems, and the potential applications of induction coupling, a system that integrates induction coupling in a unit to unit basis is certainly possible. Based on the review of current navigation systems, it seems highly possible that a homogenous system consisting of small units with induction coupling coils could be implemented. This new design could allow for an even longer operational lifetime than the University of Manchester paper, or could possibly be a viable supplement to the system outlined in that paper.

12

Works Cited

[1] F. Arvin, S. Watson, A. E. Turgut, J. Espinosa, T. Krajnik, and B. Lennox, “Perpetual Robot Swarm: Long-Term Autonomy of Mobile Robots Using On-the-fly Inductive Charging,” vol. 92, no. 3–4, pp. 395–412, 2018. [2] D. Q. Bao and I. Zelinka, “Obstacle Avoidance for Swarm Robot Based on Self-Organizing Migrating Algorithm,” vol. 150, pp. 425–432, Jan. 2019. [3] S. Kuswadi, S. I. Adji, R. Sigit, M. N. Tamara, and M. Nuh, “Disaster swarm robot development: On going project.” pp. 45–50, 2017. [4] M. Patil, T. Abukhalil, S. Patel, and T. Sobh, “UB robot swarm — Design, implementation, and power management.” pp. 577–582, 2016. [5] M. P. Vicmudo, E. P. Dadios, and R. R. P. Vicerra, “Path planning of underwater swarm robots using genetic algorithm.” pp. 1–5, 2014. [6] M. P. Vicmudo and E. P. Dadios, “Artificial neural network controller for maintaining underwater swarm robots’ wireless connections.” pp. 1–6, 2015. [7] K. J. Miraswan and N. U. Maulidevi, “Particle swarm optimization and fuzzy logic control in gas leakage detector mobile robot.” pp. 150–155, 2015. [8] Husnawati, G. F. Fitriana, and S. Nurmaini, “The development of hybrid methods in simple swarm robots for gas leak localization.” pp. 197–202, 2017. [9] X. Peng, S. Zhang, and Y. Huang, “Pattern formation in constrained environments: A swarm robot target trapping method.” pp. 455–460, 2016.

13

Annotated Bibliography [1] F. Arvin, S. Watson, A. E. Turgut, J. Espinosa, T. Krajnik, and B. Lennox, “Perpetual Robot Swarm: Long-Term Autonomy of Mobile Robots Using On-the-fly Inductive Charging,” vol. 92, no. 3–4, pp. 395–412, 2018. This article discusses the issue of small battery capacity is s...