- Shopping Bag ( 0 items )
The definitive survey of computational intelligence from luminaries in the field
Computational intelligence is a fast-moving, multidisciplinary field - the nexus of diverse technical interest areas that include neural networks, fuzzy logic, and evolutionary computation. Keeping up with computational intelligence means understanding how it relates to an ever-expanding range of applications. This is the book that ties it all together - and puts that understanding well within your reach.
In Computational Intelligence: The Experts Speak, editors David B. Fogel and Charles J. Robinson present an unmatched compilation of expanded papers from plenary and special lecturers attending the 2002 IEEE World Congress on Computational Intelligence. Collectively, these papers provide a compelling snapshot of the issues that define the industry, as observed by some of the top minds in the computational intelligence community. In a series of topical chapters, this comprehensive volume shows how current technology is shaping computational intelligence, and it delivers eye-opening insights into the field's future challenges.
The research detailed here covers an array of leading-edge applications, from coevolutionary robotics to underwater sensors and cognitive science, in such areas as:
Whatever your role might be in this dynamic, influential field, this is the one reference that no practitioner of computational intelligence should be without.
JORDAN B. POLLACK, HOD LIPSON, PABLO FUNES, and GREGORY HORNBY
The field of robotics today faces a practical economic problem: flexible machines with minds cost so much more than manual machines and their humans operators. Few would spend $2000 on a vacuum cleaner when a manual one is $200, or half a million dollars on a driverless car when a regular car is $20,000, plus $6 per hour for its driver. The high costs associated with designing, building, and controlling robots have led to a stasis, and robots in industry are only applied to simple and highly repetitive manufacturing tasks. Even though sophisticated teleoperated machines with sensors and actuators have found important applications (exploration of inaccessible environments, for example), they leave very little decision, if at all, to the on-board software.
The central issue addressed by our work is a low-cost way to get a higher level of complex physicality under control. We seek more controlled and moving mechanical parts, more sensors, more nonlinear interacting degrees of freedom, without entailing both the huge fixed costs of human design and programming, and the variable costs in manufacture and operation. We suggest thatthis can be achieved only when robot design and construction are fully automatic, and the results are inexpensive enough to be disposable.
Traditionally, robots are designed on a disciplinary basis: mechanical engineers design complex articulated bodies, with state-of-the-art sensors, actuators, and multiple degrees of freedom. These elaborate machines are then thrown over the wall to the control department, where software programmers and control engineers struggle to design a suitable controller. Even if an intelligent human can learn to control such a device, it does not follow that automatic autonomous control can be had at any price. Humans drastically underestimate animal brains: looking into nature, we see animal brains of very high complexity, brains more complex than the bodies they inhabit, controlling bodies that have been selected by evolution precisely because they were controllable by those brains. In nature, the body and brain of a horse are coupled tightly, the fruit of a long series of small mutual adaptations; like chicken and egg, neither one was designed first. There is never a situation in which the hardware has no software, or where a growth or mutation-beyond the adaptive ability of the brain-survives. The key is thus to evolve both the brain and the body, simultaneously and continuously, from a simple controllable mechanism to one of sufficient complexity for a particular specialized task.
The focus of our research is how to automate the integrated design of bodies and brains using a coevolutionary learning approach. The key is to evolve both the brain and the body simultaneously from a simple controllable mechanism to one of sufficient complexity for a task. Within a decade we see three technologies that are maturing past the threshold to make this possible. One is the increasing fidelity of advanced mechanical design simulation, stimulated by profits from successful software competition. The second is rapid, one-off prototyping and manufacture, which is proceeding from three-dimensional (3D) plastic layering to stronger composite and metal (sintering) technology. The third is our understanding of coevolutionary machine learning in the design and intelligent control of complex systems.
Coevolution, when successful, dynamically creates a series of learning environments each slightly more complex than the last, and a series of learners that are tuned to adapt in those environments. Sims' work on body-brain coevolution and the more recent Framsticks simulator demonstrated that the neural controllers and simulated bodies could be coevolved. The goal of our research in coevolutionary robotics is to replicate and extend results from virtual simulations such as these to the reality of computer-designed and constructed special-purpose machines that can adapt to real environments. We are working on coevolutionary algorithms to develop control programs that operate realistic physical-device simulators, both commercial-off-the-shelf and our own custom simulators, where we finish the evolution inside real embodied robots. We are interested ultimately in mechanical structures that have complex physicality of more degrees of freedom than anything that has ever been controlled by human-designed algorithms, with lower engineering costs than currently possible because of minimal human-design involvement in the product.
It is not feasible that controllers for complete structures could be evolved (in simulation or otherwise) without first evolving controllers for simpler constructions. Compared to the traditional form of evolutionary robotics, which serially downloads controllers into a piece of hardware, it is relatively easy to explore the space of body constructions in simulation. Realistic simulation is also crucial for providing a rich and nonlinear universe. However, while simulation creates the ability to explore the space of constructions far faster than real-world building and evaluation could, there remains the problem of transfer to real constructions and scaling to the high complexities used for real-world designs.
1.3 RESEARCH THRUSTS
We describe three major thrusts in achieving fully automated design (FAD) and manufacture of high-part-count autonomous robots. The first is evolution inside simulation, but in simulations more and more realistic, so the results are not simply visually believable, as in Sims' work, but also buildable. We investigated transferring evolved high part-count, static structures from simulation to the real world. The second thrust is to evolve automatically buildable dynamic machines that are nearly autonomous in both their design and manufacture. The third thrust, and perhaps hardest, addresses scaling to more complex tasks: handling complex, high part-count structures through modularity. We have preliminary and promising results in each of these areas, which we outline below.
1.3.1 Buildable Simulation
Commercial computer-aided design (CAD) models are in fact not constrained enough to be buildable, because they assume a human provides numerous constraints to describe reality. In order to evolve both the morphology and behavior of autonomous mechanical devices that can be built, one must have a simulator that operates under many constraints, and a resultant controller that is adaptive enough to cover the gap between the simulated and real world. Features of a simulator for evolving morphology are:
Representation: should cover a universal space of mechanisms. Conservative: because simulation is never perfect, it should preserve a margin of safety. Efficient: it should be quicker to test in simulation than through physical production and test. Buildable: results should be convertible from a simulation to a real object.
One approach is to custom build a simulator for modular robotic components, and then evolve either centralized or distributed controllers for them. In advance of a modular simulator with dynamics, we recently built a simulator for (static) Lego bricks, and used very simple evolutionary algorithms to create complex Lego structures, which were then constructed manually. Our model considers the union between two bricks as a rigid joint between the centers of mass of each one, located at the center of the actual area of contact between them. This joint has a measurable torque capacity. That is, more than a certain amount of force applied at a certain distance from the joint will break the two bricks apart. The fundamental assumption of our model is this idealization of the union of two Lego bricks. The genetic algorithm reliably builds structures that meet simple fitness goals, exploiting physical properties implicit in the simulation. Building the results of the evolutionary simulation (by hand) demonstrated the power and possibility of fully automated design. The long bridge of Figure 1.1 shows that our simple system discovered the cantilever, while the weight-carrying crane shows it discovered the basic triangular support.
1.3.2 Evolution and Construction of Electromechanical Systems
The next step is to add dynamics to modular buildable physical components, and to insert their manufacturing constraints into the evolutionary process. We are experimenting with a new process in which both robot morphology and control evolve in simulation and then replicate automatically into reality. The robots comprise only linear actuators and sigmoidal control neurons embodied in an arbitrary thermoplastic body. The entire configuration is evolved for a particular task and selected individuals are printed preassembled (except motors) using 3D solid printing (rapid prototyping) technology, later to be recycled into different forms. In doing so, we establish for the first time a complete physical evolution cycle. In this project, the evolutionary design approach assumes two main principles: (1) to minimize inductive bias, we must strive to use the lowest-level building blocks possible, and (2) we coevolve the body and the control, so that they stimulate and constrain each other. We use arbitrary networks of linear actuators and bars for the morphology, and arbitrary networks of sigmoidal neurons for the control. Evolution is simulated starting with a soup of disconnected elements and continues over hundreds of generations of hundreds of machines, until creatures that are sufficiently proficient at the given task emerge. The simulator used in this research is based on quasi-static motion. The basic principle is that motion is broken down into a series of statically stable frames solved independently. While quasi-static motion cannot describe high-momentum behavior such as jumping, it can accurately and rapidly simulate low-momentum motion. This kind of motion is sufficiently rich for the purpose of the experiment and, moreover, it is simple to induce in reality since all real-time control issues are eliminated. Several evolution runs were carried out for the task of locomotion. Fitness was awarded to machines according to the absolute average distance traveled over a specified period of neural activation. The evolved robots exhibited various methods of locomotion, including crawling, ratcheting, and some forms of pedalism (Fig. 1.2). Selected robots are then replicated into reality: their bodies are first fleshed to accommodate motors and joints, and then copied into material using rapid prototyping technology. A temperature-controlled print head extrudes thermoplastic material layer by layer, so that the arbitrarily evolved morphology emerges preassembled as a solid 3D structure without tooling or human intervention. Motors are then snapped in, and the evolved neural network is activated (Fig. 1.3). The robots then perform in reality as they did in simulation.
1.3.3 Modularity Through Generative Encodings
The main difficulty for the use of evolutionary computation for design is that it is doubtful whether it will reach the high complexities necessary for practical engineering. Since the search space grows exponentially with the size of the problem, search algorithms that use a direct encoding for designs will not scale to large designs. An alternative to a direct encoding is a generative specification, which is a grammatical encoding that specifies how to construct a design. Similar to a computer program, a generative specification can allow the definition of reusable subprocedures allowing the design system to scale to more complex designs than can be achieved with a direct encoding. Ideally, an automated design system would start with a library of basic parts and would iteratively create new, more complex modules from ones already in its library. The principle of modularity is well accepted as a general characteristic of design, as it typically promotes decoupling and reduces complexity. In contrast to a design in which every component is unique, a design built with a library of standard modules is more robust and more adaptable. Our system for automated modular design uses Lindenmayer systems (L-systems) as the genotype evolved by the evolutionary algorithm. L-systems are a grammatical rewriting system introduced to model the biological development of multicellular organisms. Rules are applied in parallel to all characters in the string, just as cell divisions happen in parallel in multicellular organisms. Complex objects are created by successively replacing parts of a simple object by using the set of rewriting rules. Using this system we have evolved 3D static structures, and locomoting mechanisms, some of which are shown in Figure 1.4, and transferred successfully into reality, as seen in Figure 1.5.
Can evolutionary and coevolutionary techniques be applied to real physical systems? In this chapter we have presented a selection of our work, each of which addresses a physical evolutionary substrate in one or more dimensions. We have overviewed research in the use of simulations for handling high-part-count static structures that are buildable, dynamic electromechanical systems with complex morphology that can be built automatically, and generative encodings as a means for scaling to complex structures. Our long-term vision is that both the morphology and control programs for robots arise directly through morphology and control-software coevolution: starting from primitive controllers attached to primitive bodies, the evolutionary system scales to complex, modular creatures by increasing the dictionary of components as stored in the creature encoding. Our current research moves toward the overall goal down multiple interacting paths, where what we learn in one thrust aids the others. We envision the improvement of our hardware-based evolution structures, expanding focus from static buildable structures to buildable robots. We see a path from evolution inside CAD/CAM (computer-aided manufacture) and buildable simulation, to rapid automatic construction of novel controlled mechanisms, and finally the use of generative encodings to achieve highly complex, modular individuals. We believe such a broad program is the best way to ultimately construct complex autonomous robots whose corporate assemblages consist of simpler, automatically manufactured parts.
This research was supported in part by the National Science Foundation (NSF), the office of Naval Research (ONR), and the Defense Advanced Research Projects Agency (DARPA).
Excerpted from Computational Intelligence Copyright © 2003 by IEEE. Excerpted by permission.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
1. THREE GENERATIONS OF COEVOLUTIONARY ROBOTICS (Jordan B. Pollack, Hod Lipson, Pablo Funes, and Gregory Hornby).
1.3 Research Thrusts.
2. BEYOND 2001: THE LINGUISTIC SPATIAL ODYSSEY (James M. Keller, Pascal Matsakis, and Marjorie Skubic).
2.2 Force Histograms and Linguistic Scene Description.
2.3 Scene Matching.
2.4 Human–Robot Dialog.
2.5 Sketched Route Map Understanding.
2.6 The Future.
3. COMPUTING MACHINERY AND INTELLIGENCE AMPLIFICATION (Steven K. Rogers, Matthew Kabrisky, Kenneth Bauer, and Mark E. Oxley).
3.2 Estimating Intelligence.
3.3 Turing Test and Intelligence Amplification.
3.4 Measuring Intelligence Amplification.
3.5 The Future of Intelligence Amplification.
4. VISUALIZING COMPLEXITY IN THE BRAIN (Lloyd Watts).
4.2 Neuroscience Knowledge.
4.3 Computing Technology.
4.4 Nontechnical Issues.
5. EMERGING TECHNOLOGIES: ONR’S NEED FOR INTELLIGENT COMPUTATION IN UNDERWATER SENSORS (James F. McEachern and Robert T. Miyamoto).
5.3 The Challenge.
5.4 Current Applications.
6. BEYOND VOLTERRA AND WIENER: OPTIMAL MODELING OF NONLINEAR DYNAMICAL SYSTEMS IN A NEURAL SPACE FOR APPLICATIONS IN COMPUTATIONAL INTELLIGENCE (Rui J. P. de Figueiredo).
6.2 Classes of Nonlinear Dynamical System Models.
6.3 The de Figueiredo–Dwyer–Zyla Space F.
6.4 Derivation of Sigmoid Functionals.
6.5 Best Robust Approximation of f in the Neural Space N.
6.6 Optimal Combined Structural and Parametric Modeling of Nonlinear Dynamical Systems in N.
6.7 Computationally Intelligent (CI) Systems.
6.8 Concluding Remarks.
7. TECHNIQUES FOR EXTRACTING CLASSIFICATION AND REGRESSION RULES FROM ARTIFICIAL NEURAL NETWORKS (Rudy Setiono).
7.2 Rule Extraction.
7.3 Illustrative Examples.
8. NEURAL NETWORKS FOR CONTROL: RESEARCH OPPORTUNITIES AND RECENT DEVELOPMENTS (Paul J. Werbos).
8.1 The Challenge to Researchers: Context and Motivation.
8.2 A Specific Challenge and Associated Issues.
8.3 Strategies, Tasks, and Tools.
9. INTELLIGENT LEARNING ROBOTIC SYSTEMS USING COMPUTATIONAL INTELLIGENCE (Toshio Fukuda and Naoyuki Kubota).
9.2 Motion Planning and Behavior Acquisition of Robots.
9.3 Emerging Synthesis of Computational Intelligence.
9.4 Intelligence on Robotic Systems.
9.5 Structured Intelligence for Robotic Systems.
9.6 Concluding Remarks.
10. COMPUTATIONAL INTELLIGENCE IN LOGISTICS (Hans-Jürgen Zimmermann).
10.2 Traffic Management.
10.3 Fleet Management.
10.4 In-House Logistics.
11. TWO NEW CONVERGENCE RESULTS FOR ALTERNATING OPTIMIZATION (James C. Bezdek and Richard J. Hathaway).
11.1 Iterative Optimization.
11.2 Existence and Uniqueness.
11.3 The Alternating Optimization Algorithm.
11.4 When Is Alternating Optimization a Good Choice?
11.5 How Do We Solve (11.1)?
11.6 Local Convergence of Alternating Optimization.
11.7 Global Convergence of AO.
12. CONSTRUCTIVE DESIGN OF A DISCRETE-TIME FUZZY CONTROLLER BASED ON PIECEWISE-LYAPUNOV FUNCTIONS (Gang Feng, Dong Sun, and Louis Wang).
12.2 Fuzzy Dynamic Model and Its Piecewise-Quadratic Stability.
12.3 Controller Synthesis of Fuzzy Dynamic Systems.
12.4 Simulation Examples.
13. EVOLUTIONARY COMPUTATION AND COGNITIVE SCIENCE (Janet Wiles and Jennifer Hallinan).
13.1 Cognitive Science: What’s on Your Mind?
13.2 Case Studies in Evolutionary Computation and Cognitive Science.
14. EVOLVABLE HARDWARE AND ITS APPLICATIONS (T. Higuchi, E. Takahashi, Y. Kasai, T. Itatani, M. Iwata, H. Sakanashi, M. Murakawa, I. Kajitani, and H. Nosato).
14.2 Myoelectric Prosthetic Hand Controller with EHW.
14.3 Data-Compression Chip for Printing Image Data.
14.4 Analog EHW Chip for Cellular Phone.
14.5 An EHW-Based Clock-Timing Adjusting Chip.
14.6 Evolvable Optical Systems and Their Application.
14.7 Current Research on EHW.
15. HUMANIZED COMPUTATIONAL INTELLIGENCE WITH INTERACTIVE EVOLUTIONARY COMPUTATION (Hideyuki Takagi).
15.2 Humanized Computational Intelligence.
15.3 Interactive Evolutionary Computation.
16. UNSUPERVISED LEARNING BY ARTIFICIAL NEURAL NETWORKS (Harold Szu).
16.1 A New Challenge: Space-Variant Unsupervised Classifications.
16.2 Power of Pairs: Vector versus Scalar Data.
16.3 Generalization of Shannon’s Entropy Information Theory to Open Systems.
16.4 Benchmarks of Space-Variant Unsupervised Classification.
16.5 Multispectral Medical Imaging.
16.6 Multispectral Remote Sensing.
16.7 Biological Relevance.
17. COLLECTIVE INTELLIGENCE (David H. Wolpert).
17.1 Motivation and Background.
17.2 The Mathematics of Designing Collectives.
17.3 Tests of the Mathematics.
18. BACKPROPAGATION: GENERAL PRINCIPLES AND ISSUES FOR BIOLOGY (Paul J. Werbos).
18.2 The Chain Rule for Ordered Derivatives.
18.3 Backpropagation for Supervised Learning.
18.4 Discussion and Future Research.
ABOUT THE EDITORS.
Posted June 5, 2013