Dance is the most perishable of the arts. Traditionally, dances are taught by one generation of dancers to the next; if memory fails or the chain of transmission breaks, then the dance is lost. Recent works of major significance have been lost in this way. “The Seven Deadly Sins” was choreographed in 1933 by George Balanchine to music by Kurt Weill and a libretto by Bertolt Brecht; the music and libretto of course survive, but the original choreography is gone with no hope of recovery. Similarly, the folk dances of minority cultures are under threat of disappearance in the same way as their arts, language, and music, and the dances are much harder to record and preserve.

Video recordings are of course enormously helpful, but by universal consensus inadequate. Notations for dance analogous to musical notation have been developed. Probably the best known, and certainly the most extensively discussed in this book, is Labanotation, invented by Rudolf von Laban (1879-1958), but at least 80 others have been proposed. Dances can indeed be recorded and then reconstructed from such notations, particularly when supplemented by video. However, the notation is very difficult to learn and to use. What must be recorded--the combined trajectories of a company of human bodies over time and their relations to one another and to the music--is inherently exceedingly complex. As a result, reading dance notations is difficult, and writing dance notation is extremely difficult, and very slow and laborious.

An analogous problem arises in robotics and artificial intelligence. It is reasonably straightforward to characterize robotic actions at either a very low level, in terms of the angle of joints and similar geometric parameters, or at a very high level in terms of abstract effects (for example, ”Put the red block on top of the blue block”). Theories of planning exist at both these levels; the first is surveyed in [1], the second in [2]. However, finding a systematic representation for robotic actions at a middle level of abstraction to bridge the gap between these has proved to be very difficult and is a major challenge in theories of automated planning. (In these terms, Labanotation is comparatively low-level, though imprecise as compared to a robotic control language.)

This book comes out of a multi-disciplinary workshop of the same title held at LAAS-CNRS in 2014. The participants include dance theorists, roboticists, and computer scientists, mostly from France and Japan, but also from Austria, Canada, Germany, the Netherlands, and the US. The 20 papers discuss a wide range of related issues: for instance, software for dance notation, facilitating the recording, editing, and reading of dance notation; a project to record traditional Taiwanese folk dances; issues in creating animations for sign language and theatrical gestures; techniques for getting robots to imitate human motion (difficult, because in general robots are much less flexible); a taxonomy of the kinds of grasps used in everyday actions; and a geometric and physical analysis of the movements involved in dancing a tango, using motion-capture technology. Curiously, as far as I could tell, the chapter on the tango was the only one to consider problems involved with multiple dancers; the other 19 chapters consider only a single dancer.

As is common in this kind of collection, the whole is hardly more than the sum of its parts. I found several of the chapters very interesting and several more worth reading, but overall the collection gives a rather fragmentary view of the field. If someone were to write a paper or monograph that surveyed this material systematically and placed it in a common framework, that would be a major contribution.

Thanks to Andrew Sundstrom and Tiffany Mills for helpful information.