1R14. Adaptive Neural Control of Walking Robots. Engineering Research Series, Vol 5. - MJ Randall (Deceased). Professional Eng Publ, Suffolk, UK. 2001. 332 pp. ISBN 1-86058-294-X. $69.00.
Reviewed by JE Cochran (Dept of Aerospace Eng, Auburn Univ, 211 Aerospace Eng Bldg, Auburn AL 36849-5338).
Scientists and engineers have often used nature as a source of both inspiration and practical solutions in the design of machines. It follows that when an engineer considers the problem of designing a walking machine, it can be expected that he or she will try models of walking animals. As a basis for the design of a walking machine capable of transversing rough terrain, the author of this monograph chose to study the walking processes of hexapod insects. On the basis of observations of these animals and the application of mathematical tools such as neural networks and optimal control, he developed both the theory and a working model of a walking robot that was to some degree successful. The author dedicated much of his unfortunately short life to the work described in this book, and it is a lasting tribute to him and to his professors and co-workers who are responsible for its publication.
This book should be of interest to everyone involved in the design of walking robots, to engineers and scientists interested in the application of neural networks and optimal control to electromechanical systems, and even to professors teaching philosophy of science who have good mathematical backgrounds. The extensive coverage of the literature relating to robots and the use of models of insect motion alone makes this book a valuable resource. It is not a textbook, but could be used in a graduate course as one of several resources.
The author considers many areas ranging from philosophy to applied psychology, to biology, to dynamics, and to insect neuro-physiology. Starting with Aristotle, the author provides, in Chapter 1, background on walking robots, gives many uses for such machines, and discusses several operational robots such as the Plustech Forestry Harvester and the Honda bipeds. The author approaches the problem of designing a walking robot by developing, in Chapter 2, a novel generic control hierarchy consisting of “four layers:” motivation, body route trajectory generation, kinematic planning, and dynamic execution. In Chapter 3, he considers the emulation of walking strategies of insects that transverse rough terrain to obtain basic principles regarding sensing and control. Walking animals exhibit gaits, ie, repetitive motions of their limbs, which are used for slow and fast motion.
Chapter 4 deals with gaits from the standpoint of biological oscillatory behavior, and the research on stability of interconnected oscillators is reviewed. Various approaches to robot leg trajectory generation are considered in Chapter 5. These include the top-down and bottom-up approaches to motion generation. In the top-down approach, the motion is observed, and a mathematical model of the motion is used to determine the forces and moments needed to generate it. The top-down approach is similar to dynamic inversion used in the control of aerospace flight vehicles. An interesting concept, the isochrony principle, which requires that a trajectory segmented by way points be generated so that the times between way points are equal, is used to generate velocity and acceleration time histories. The bottom-up approach to motion generation is based on the notion that the motion should be such that the time rates of change of the linear and angular accelerations (the linear and angular jerks) are minimized. This requires the use of optimal control principles.
In Chapter 6, the author considers the kinematics and dynamics (kinetics?) of hexapods. A closed-form solution for the forward and inverse kinematics of a single leg is derived. A review of the theory of adaptive neural control is presented in Chapter 7. Controllers classified as linear-equivalent and nonlinear equivalent are described, and the former type is chosen. The synthesis of stable adaptive neural controllers for open-chain dynamic systems is addressed using Lyapunov’s Second Method. Some experiments are also discussed. Proofs of certain theorems are given in the Annex to this chapter.
Since in walking the hexapod robot produces closed, rather than open chains when two or more of its feet are on the ground, the stability of closed kinematic chains is considered in Chapter 8, and the most important results of the author’s work are presented in the form of four theorems. As the author writes in the concluding chapter (Ch 10), “…the stability proof does not guarantee that the robot will not fall over… However, it does guarantee that if such a failure occurs, it would not be as a result of the instability of the neural controllers.”
Preliminary experimental results are presented in Chapter 9. These show the promise of the approach taken by the author. In Chapter 10, he suggests that his work could be extended at both the theoretical and practical levels. Interested graduate students and other researchers will find some challenging ideas there.
As noted above, individual researchers in the area of robotics will find Adaptive Neural Control of Walking Robots of interest. It is also recommended as a selection for libraries at institutions where research in this area is conducted.