Massachusetts Institute of Technology 16.412J/6.834J Cognitive Robotics Problem Set 1 Cognitive Robot Types, Topics and Reasoning Methods Lawrence Bush February 15, 2005 Part A: Topics of Fascination In this section I will discuss the following three topics of interest in the area of reasoning applied to robotics: Collaborative Task Execution Resource Allocation Learning Collaborative Task Execution and Resource Allocation Collaborative Task Execution and Resource Allocation are closely related. Therefore, I will address them together. The objective of collaboration is to produce a higher quality result than could be achieved alone. The objective of resource allocation is to increase efficiency or lower costs. These two concepts embody the quality – cost tradeoff of many operations research optimization problems. The central idea is that teams or colonies of robots would need to be able to self organize in a way that allocates them in an efficient manner. The tasks that need to be done would change as time goes on. For a given task, the robots need to decide who does what. For example, in a search and rescue operation, the robots need to spread out over an area to search for the missing person. The way in which the robots are deployed will significantly affect the success of the mission. Suppose that prior knowledge indicates that the missing person is more likely to be in one location over another. Also, certain areas are easier (lower cost) to search than others. Consequently, if each robot operates independently, they may all choose the area that is easiest to search and has the highest probability of success while neglecting other areas. A more optimal solution can be found. Therefore, another way to collaborate is desirable for this application. The robot allocation could be centrally planned and modeled as a Markov Decision Process (MDP) or constraint satisfaction problem or it could be constructed using a market-based approach. These are very interesting approaches and will be discussed more in Part D. Learning Learning is a very broad topic in artificial intelligence. The index of Artificial Intelligence, A Modern Approach 2nd Edition, by Stuart Russell and Peter Norvig (AIMA) lists over 100 pages in the book that address the topic of learning. The broad reaching nature of the topic is captured in titles like “universal 1reinforcement learner.” While that goal seems intractable, robots still need to be able to perceive reason and act in new and changing environments. A variety of learning methods have been studied for many applications. Traditional learning algorithms have a narrow application scope. These approaches are very interesting, important and challenging. However, robots need to learn from experience in a more general sense. They need to learn to identify what is important, internalize this into their knowledge base and use it to make decisions in the future. Two approaches to this problem are recurrent neural networks (http://www.idsia.ch/~juergen/rnn.html) and optimal ordered problem solver (http://www.idsia.ch/~juergen/oops.html). The referenced web site is flat out fascinating. With that said, I find this reasoning area to be overly ambitious. My approach to a problem this big is to nibble away at the edges. In short, I intend to choose a problem that is more specific and narrower in scope. A Summary of Robotics This section provides an overall view of robotics. I wrote this section in order to become familiar with the different types of robots, sensors, reasoning techniques and applications in the field. This section essentially summarizes Chapter 25 of AIMA. This section is organized in the following order: • Sensors • Reasoning • Types of Robots • Applications A robot is a feedback and control system that involves sensing and reasoning about its environments, which leads to a decision on how to act. The first three parts of this section are organized in the order in which they are used in a feedback and control system. The final part covers how robots can be applied. Sensors There are many kinds of sensors that a robot can use [1] to sense its environment: Range Finders Laser Range Finder Sonar Global Positioning System Tactile Sensors (whiskers) Imaging Sensors Camera Infrared Other Temperature Odor Acoustic Velocity Light Intensity The sensors that are chosen for the particular application become part of the feedback loop described above. This feedback loop is sometimes called perception. Perception is essentially a filtering task, which 2involves sensing, updating the internal belief state and acting. For example, the important and well-studied localization problem involves sensing the robot’s location (i.e. using a range finder), updating the robot’s estimated location (i.e. Kalman Filter) and moving in the direction that leads to the goal. Proprioceptive Sensors are another class of sensors that measure the internal state of the robot. Some examples of these are: • Inertial (gyroscope) • Force / torque sensors • Odometer While these measure the internal state of the robot, they can also be used to improve the robot’s interpretation of its environment. For example, an inertial sensor will reduce the position uncertainty in a localization task. Reasoning After sensing the environment, a robot must update its internal state and make a decision about how to act. This reasoning step can have a variety of objectives, for example: • Localization • Mapping • Task Planning • Resource Allocation • Generic Search • Learning (in general) Acting (Types of Robots) The manifestation of the acting step depends on the type of robot and how it can affect its environment. Robots can be grouped into 3 types: manipulators, mobile and hybrid robots. Manipulators are able to change their environment through grasping, pushing, lifting etcetera. An assembly line robot is a popular example of this type of robot. A mobile robot is able to move about via wheels or legs. Some examples of mobile robots are unmanned land vehicle, unmanned air vehicle, autonomous underwater vehicles, and planetary rovers. A hybrid robot is a mobile robot equipped with manipulators, for example a humanoid robot. Applications A robot’s ability to sense, reason and interact with its environment is application dependent. Robots are used in industry, exploration, health care and personal services.
View Full Document
Unlocking...