Keywords
Citation
Andrew, A.M. (2000), "Mobile Robotics: A Practical Introduction", Kybernetes, Vol. 29 No. 4, pp. 523-529. https://doi.org/10.1108/k.2000.29.4.523.3
Publisher
:Emerald Group Publishing Limited
The innovative nature of the contents of this little book is not immediately apparent from the title nor from a quick flick through the pages. There are pictures of a commercially‐available “Nomad” robot, and of other self‐propelled wheeled devices with which experiments have been conducted. A legged‐locomotion device is mentioned, as are planetary rovers and devices for underwater exploration, etc., but only in passing. The hardware illustrations suggest a pedestrian treatment (a rather strange suggestion to be supported by lack of attention to legged locomotion!) and this is reinforced by the author’s careful development of basic topics, with discussion of the nature and characteristics of alternative types of sensors and actuators, and of statistical techniques used to assess performance.
In fact, however, the focus of the book is very far from being pedestrian (in the metaphorical sense) and it contributes, as the notes on the back cover claim, to a cutting‐edge research topic. It is a pity that the title does not hint at this by the inclusion of some such term as “adaptive” or “viable”. The book introduces a new approach to mobile robotics aimed at the achievement of animal‐like robustness and versatility. This is contrasted with earlier work in which the computer control was based on AI principles and was considered separately from what were seen as relatively uninteresting engineering aspects of sensors and actuators. The older approach is summed up by the equation:(see equation 1)The use of classical AI methods requires symbols whose relevance to the “real world” stems from the insights of the programmer, and the operation is “brittle” because it is impossible to foresee every eventuality. In the new approach, sensors and actuators are integrated more closely, without dependence on arbitrary symbols (even though a symbolic explanation may be apparent in retrospect). It is acknowledged that this tighter linking of perception and action must be at the expense of complexity, but with the hope that the “intelligent bit” will arise as a result of interaction between various relatively simple processes. This would be an “emergent phenomenon” or “synergetic effect”.
Ways in which a mobile robot can be controlled, with a minimum of built‐in assumptions about the nature of the environment, are illustrated by 12 “case studies” which are experiments conducted by the author and colleagues, mainly using the Nomad robot mentioned earlier. The latter is equipped with several kinds of proximity sensor on each of its 16 faces, and for no obvious reason is given the name “FortyTwo”. It has a television camera and an on‐board computer and the possibility of a radio link to a more powerful computer.
The case studies show ways of achieving an impressive repertoire of kinds of learned behaviour, navigation and map‐building. They are interesting not only from the point of view of robotics applications but also for their comparison with similar performance in animals. They employ an intriguing range of biologically‐inspired techniques, including neural nets of the multi‐layer perceptron type, especially a variant used as a pattern associator, and also of the self‐organising feature map (SOFM) due to Kohonen. There is also application of the “adaptive resonance theory” (ART) associated with the pioneer of neuromodelling, Stephen Grossberg. The ART has been advanced as a theory of perception and classification in biological systems and this is a valuable exploration of its practical application. Frequent reference is made to the construction of a cognitive map in the hippocampus of the animal nervous system.
In each of the case studies the robot showed useful learning in an impressively small number of trials. The first studies were made with the robot initially under manual control, but with learning governed by simple rules referred to as “instincts” and leading quickly to useful patterns of autonomous action. Another study uses complex input data, from an omnidirectional camera, for perceptual localisation.
In another study the record of robot actions, corresponding to proprioceptive information in a nervous system, is used to establish location along a path. It is particularly interesting that this is done using Kohonen’s SOFMs, but with a range of SOFMs operating in parallel with and sensitive to different lengths of record. A great deal of new ground is explored here in ways of utilising these various adaptive schemes.
As mentioned earlier, the author develops his topics with care, and the material is presented in textbook fashion, including exercises for the student. Although the use of computer simulation is discussed, it is argued that simulation is never exact and there is no substitute for working with a real robot, and the student is strongly advised to acquire one. There is a large and useful bibliography, and a sprinkling of relevant Web sites.
The book is an admirable introduction to this modern approach to mobile robotics and certainly gives a great deal of food for thought. At present the concrete results are limited to navigation and simple tasks of box‐pushing and floor cleaning, but it is easy to feel that there is the potential for very much more and that we are being given a glimpse of principles for future animal‐like and humanoid robots. Also, the vigorous exploration of the possibilities of using neural nets and other biologically‐inspired techniques in real‐world control situations must have implications for theories of biological evolution. This is an important and thought‐provoking book.