《智能移动机器人》结课论文

华中科技大学研究生课程考试答题本考生姓名考生学号系、年级类别考试科目智能移动机器人考试日期 2014年12月26日评分注意：1、无评卷人签名试卷无效2、必须用钢笔或者圆珠笔阅卷，使用红色。

用铅笔阅卷无效（双面打印）Control Architectures for IntelligentMobile RobotsAbstract. Control architecture is a very important concept for the designof intelligent mobile robots. This article has a simple introduction aboutthe dominant control architectures, i.e. deliberative control architecture,reactive control architecture hybrid deliberative/reactive controlarchitecture and behavior-based control architecture. Meanwhile,introduce the subsumption control architecture which is similar to thebehavior-based control architecture, but very popular. In the time, thearticle makes comparisons between them, and introduces theiradvantages and disadvantages.Keywords. Mobile robots, Robot control, Control architecture1 IntroductionIn the 21st century, it is expected that robots with different degrees of autonomy and mobility will play an increasingly important role in all aspects of human life. With the robotic technologies development of the robotics with the past few years, there are more and more inspiring and remarkable success stories for robotics. And mobile robots have been widely applied in many fields. Such as factory automation dangerous environment detection, office automation, hospital, entertainment, education, space exploration, farm automation, military and security system.Furthermore, mobile robots pose a unique challenge to artificial intelligence researchers. Mobile robots need to be used to deal with key issues such as uncertainty, reliability, and real-time response.For dealing with the challenge, control architecture for mobile robots becomes a growing interest and a popular subject. Robotic control architecture can be seen as a mixture of engineering and cognitive science and as such it presents unusual challenges to the programmer. Robotic control methodologies have tended to move from simplistic, predefined actuator actions based on specific input criteria to tightfeedback loops with input from the environment, giving more robust solutions. In these environments, where many such control loops are required, the opportunities for the application of parallel programming to create simple and robust solutions are numerous.Even simple robots can have many different tasks to do simultaneously. For example, a robot might try to avoid bumping into walls whilst also trying to create the largest treacle pudding in the world. The latter task is the main purpose of the robot, but the first is important for the robot to meet its goal as designed and needs to be handled constantly alongside the robot’s main task.So an excellent control architecture for mobile robots is a very important. The article has a simple introduction to the control architecture.2 The control architecture for intelligent mobile robotsIn the charter, we will have a quick look at some main control architectures which are deliberative control reactive control hybrid control and behavior-based control and introduce how they work. The Table 1 shows the difference between them.Table 1: Difference between control architectures2.1 The Deliberative Control ArchitectureThe deliberative control architecture, namely hierarchical control architecture, uses a centralized world model for verifying sensory information and generatingactions in the world. More simply, the robot takes all of the available sensory information, and all of the internally stored knowledge it has, and it reasons about it in order to create a plan of action. When it faces with the relative situation, the robot senses its environment, plans its next action based on these senses, and then takes appropriate action using available actuators.From what said above, it’s easy to find that th e robot explicitly plans its next action from the knowledge it has gathered about the environment so far at every stage. Essentially these robots just select actions from rule matches on the current ‘perceptions’ from sensory input. The follow Figure 1 shows the structure of a deliberative robot control architecture. From the figure, it’s easy to know that the work method is that the modules that be designed in advance work to read values from the sensors available to provide data about the environment, devise strategies to perform the desired behaviors given the environmental state, and then compose the signals that control the actuators to achieve those behaviors.Figure 1: The typical structure of a deliberative robot control system But in the early stage, the approach employs a top-down analysis of the desired behavior of the robot during the design phase and then the implementation of a sequence of modules. So deliberative approaches have been criticized for scaling poorly with the complexity of the problem and consequently not allowing for reaction in real-time.2.2 The Reactive Control ArchitectureStrictly speaking, the reactive control architecture is not a control architecture, for the purely reactive systems do not use any internal representation of the environment, and perform no search: they react to the current sensory information.The typical structure of a reactive robot control system (as shown in Figure 2) is very simple.Figure 2: The typical structure of a reactive robot control system Comparing with the deliberative control architecture, it can be found that reactive systems have no the ‘Plan’ unit. Actually they use a direct mapping between sensors and actuators, and minimal, if any, state information. They consist of collection of rules that map specific situations to specific actions. Just like the knee-jerk reaction, they are based on “stimulus-response”. And it won’t be decided by your brain. So we can say the way that reactive control works is similar with the unconditioned reflex of the human. It’s a great idea, because you should do all the things by your brain, or you will be very tired. It’s the same for robots. And this allows the robot to respond very quickly to changing and unstructured environments at the same time.Every coin has two sides. It can’t be exception. The reactive control architecture also has many disadvantages. For example, it is often too difficult to split up all possible situations this way, or it may require unnecessary encoding. Limitations to this approach are that such robots, because they only look up actions for any sensory input, do not usually keep much information around, have no memory, no internalrepresentation around them, and no ability to learn over time. Due to their inability to store information dynamically, purely reactive strategies are less used, although it’s effective for a variety of problems.All in all, with more and more building blocks of a reactive system (reactive rules coupling sensors and actuators) developed, it has a bright future.2.3 The Hybrid Deliberative/Reactive control architectureThe two kinds of control architectures mentioned above both have advantages and disadvantages. So designers came up with the hybrid deliberative/reactive control architecture to combine the best of both Reactive and Deliberative approaches. Just like the human, we control our bodies by spinal cords and brains. The deliberative approach like the control by the brain, and the reactive approach like the control by spinal cords. So in the hybrid deliberative/reactive control architecture, one part of the robot’s “brain” plans, while another deals with immediate reaction, such as avoiding obstacles and staying on the road. If we just put the two parts together, it’s can’t get the result that one add one greater two. The greater challenge of this approach is not just bringing the two parts of the brain together, but also allowing them to talk to each other, and resolve conflicts between the two. So a “third” part of the robot brain is required. A s a result these systems are often called “three-layer systems”.The three-layer architecture consists of three components: a reactive feedback control mechanism, a reactive plan execution mechanism, and a mechanism for performing time-consuming deliberative computations.In different robots, the components have different names. So we just named them with the controller, the sequencer, and the deliberator.The controller consists of one or more threads of computation that implement one or more feedback control loops, tightly coupling sensors to actuators. Usually the controller contains a library of hand-crafted transfer functions. Which ones are active at any given time is determined by an external input to the controller.The sequencer's job is to select which primitive Behavior (i.e. which transfer function) the controller should use at a given time, and to supply parameters for the Behavior. By changing primitive Behaviors at strategic moments the robot can be coaxed into performing useful tasks. Of course, that the outcome of selecting a particular primitive in a particular situation might not be the intended one. So the sequencer must be able to respond conditionally to the current situation, whatever it might be.The deliberator is the locus of time-consuming computations. The deliberator can interface to the rest of the system in two different ways. It can produce plans for the sequencer to execute, or it can respond to specific queries from the sequencer. The deliberator is very smart but slow. It likes your brain.Now let’s look at the structure of the hybrid deliberative/reactive control architecture (as shown in Figure 3).Figure 3: The typical structure of a hybrid robot control system From the figure 3, we can get that the sense-plan-act (SPA) approach has a significant architectural feature. It’s that the flow of control among these components is unidirectional and linear. Information flows from sensors to world model to plan to effectors, never in the reverse direction.The hybrid deliberative/reactive control architecture is very popular. There areseveral systems which demonstrate the hybrid deliberative/reactive control approach: Arkin’s AuRA [Arkin, 1986], Gat’s Atlantis [Gat 1991], Lyon’s Planner-Reactor [Lyons & Hendriks, 1992], Georgeff’s PRS [Georgeff & Lansky 1987], etc.2.4 The Behavior-Based Control ArchitectureThis chapter will introduce the behavior-based control architecture which is one of the three dominant paradigms for robot control. The other two are reactive control and hybrid control.The behavior-based control is focused around the idea of removing centralized control structures and instead linking actions directly to changes in the input sources themselves. The behavior-based approaches are an extension of reactive systems that fall between the purely reactive and the planner-based extremes. The behavior-based strategies are strictly more powerful than purely reactive approaches since they have no fundamental limitations on internal state.A behavior-based control system has a general structure that is similar to that shown in Figure 4. It is composed by various interconnected agents forming a control system in a layered fashion. The interconnections between agents are dynamically established, and are controlled by a message passing system.As can be seen in Figure 4, there are three categories of agents. The agents in charge of acquiring information about the environment are the Sensor agents and are directly or indirectly connected to physical sensors. Those indirectly connected to physical sensors, like Agent S3 in Figure 4, just process information delivered by other agents, and are called Virtual Sensor agents. In a similar way, there are Actuator agents, which send information to physical actuators present in the system. There are also Virtual Actuator agents, which are similar to the virtual sensor agents.Finally, there is a last group of agents, called Behavior agents, which receive information from the Sensor agents, process it, and send the results to the Actuator agents or other Behavior agents. As any other agent, the Behavior agents can be activeor inactive in a given instant. In Figure 4, the agent B2 is inactive. Immediately after being activated by a message sent by agent B1, it must rearrange the connections in order to modify the overall behavior of the system.Figure 4: A behavior-based control system considered as an interconnection of agentsThe key difference between behavior-based and hybrid systems is in the way representation and time-scale are handled. Hybrid systems typically employ a low-level reactive system that functions on a short time-scale, and a high-level planner that functions on a long time-scale. The two interact through a middle layer. Consequently hybrid systems are often implemented with so-called three-layer architectures.In contrast, behavior-based systems attempt to make the representation, and thus the time-scale, of the system uniform. Behavior-based representations are parallel, distributed, and active, in order to accommodate the real-time demands of other parts of the system. Furthermore, they are implemented using the behavior structure, much like the rest of the system.2.5 The Subsumption Control ArchitectureThe subsumption control architecture is probably the most widely known mobile robot control architecture. Subsumption has been widely influential in autonomous robotics and elsewhere in real-time AI. It was introduced by Rodney Brooks and colleagues in 1986.The subsumption control architecture is a reactive robotic architecture heavily associated with behavior-based robotics.Now, let’s have a look at the difference between the traditional architectures (as shown in Figure 5) and the Brook’s subsumption archi tecture (as shown in Figure 6).Figure 5: Traditional Architectures (Linear Organization)Figure 6: Brook’s Subsumption Architecture (Parallel Organization) The subsumption approach involves building robot control systems with increasing levels of competence. Each additional level builds upon and potentially interacts with the inputs and outputs of existing, previous levels to add higher levels of competency, leaving the lower levels intact, functional and operational within the overall system. The higher levels build upon the lower levels to create more complex behaviors. The behavior of the system as a whole is the result of many interacting simple behaviors. The layers operate asynchronously. Thus, each layer is fully capable of controlling the robot by itself, albeit in a possibly simple-minded way.The main advantages of the architecture are:1) The emphasis on iterative development & testing of real-time systems in their target domain;2) The emphasis on connecting limited, task-specific perception directly to the expressed actions that require it;3) The emphasis on distributive and parallel control, thereby integrating the perception, control, and action systems in a manner similar to animals.The main disadvantages of the architecture are:1) The difficulty of designing adaptable action selection through highlydistributed system of inhibition and suppression;2) The lack of large memory and symbolic representation, which seems to restrict the architecture from understanding language;When subsumption architecture was developed, the novel setup and approach of subsumption architecture allowed it to be successful in many important domains where traditional AI had failed, namely real-time interaction with a dynamic environment. The lack of large memory storage, symbolic representations, and central control, however, places it at a disadvantage at learning complex actions, in-depth mapping, and understanding language.3 ConclusionsThis article described the main control architectures for intelligent mobile robots. It has been seen that the control architecture constrains the way an autonomous robot senses, reasons and acts, thus affecting its task performance. So having a good knowledge of control architectures is essential for the designers to design a mobile robot. At the same the article makes comparisons between them in order to understand them better.AcknowledgementsAt the end of the article, I am very grateful to my teacher Professor Luo for his careful classroom instruction. Meanwhile thanks to the pioneers of the robotics for their great contribution to the world.References[1] Karrera Amaierako Proiektua. Agent-based control architecture for a mobile robot.[2] Jonathan SIMPSON, Christian L. JACOBSEN. Mobile Robot Control The SubsumptionArchitecture and occam-pi. Communicating Process Architectures. 2006[3] Erann Gat. On Three-Layer Architectures . Artificial Intelligence and Mobile Robots. AAAIPress.[4] Arkin, R.C. Behavior-Based Robotics. Cambridge. MA: MIT press. 1998[5] Brooks, R.A. A robust layered control system for a mobile robot. IEEE Journal of Roboticsand Automation, RA-2. 1986[6] Daniel Toal, Colin Flanagan. Subsumpption architecture for the control of robots.[7] Magnus Egerstedt. Control of Autonomous Mobile Robots .[8] Omead Amidi. Integrated Mobile Robot Control . Carnegie Mellon University . 1990[9] Kirsanov Kirill . The Architecture of Robotics Control Software for HeterogeneousMobile Robots Network. 24th DAAAM International Symposium on IntelligentManufacturing and Automation, 2013[10] Jonathan H. Connell. SSS: A Hybrid Architecture Applied to Robot Navigation. theProceedings of the 1992 IEEE Conference on Robotics and Automation (ICRA-92), pp.2719-2724.[11] José Eduardo Mendonça Xavier, Hansjörg Andreas Schneebeli. A Framework based onconcurrent Object-oriented programming for Building Behavior-based Control Systems for Mobile Robots.[12] Subsumption architecture. /wiki/Subsumption_architecture[13] Behavior-based robotics. /wiki/Behavior-based_robotics[14] Hierarchical control system. /wiki/Hierarchical_control_system[15] Robotic paradigms. /wiki/Robotic_paradigms。