Living Sculpture

The Art and Science of Creating Robotic Life

Written for and published in Leonardo

I am currently involved in the construction and display of a body of work I call "Living Sculpture," which represents a series of works that attempts to bring emotional intelligence and awareness to sculptured life forms. The process of creating a living sculpture is challenging due to the complexity involved in having that sculpture integrate multiple changing inputs and react in real time. From finding appropriate materials to developing technologies for gesture, locomotion, sensory input, and behavior, countless technological and physical obstacles have to be overcome to achieve a unified sculpture. I consider my work to be an attempt to find a symbiotic balance between classical artistic expression and contemporary technologies.

Yves Bouncing Around

I try to incorporate a sense of biological design into my sculptures. When we observe a living organism, we can see detail and complexity from any scale or from any angle. I believe that a living sculpture should embrace this intricate detail as much as possible. As a design progresses from the inner basic systems to the complex outer systems, surprising and elaborate aesthetic and behavioral effects may emerge that exceed our expectations. My goal is to make my living sculptures interesting and exciting from any perspective, unlike a painting which has a back that has not been treated with the same care as the rest of the artwork. Still, hardly any of the details in my sculptures are simply for aesthetic purposes. Each line and curve should have a significant effect on the functionality and aesthetics of the piece in order to survive the design process. In a sense, I apply some of the principles of natural selection to a sculpture in progress.

Fig. 1. The Scorpibot (1992), a 6-foot-tall mobile sculpture with a black rubbery skin, a fiberglass shell, and treads for locomotion. Scorpibot receives information about the outside world from infared and touch sensors; the hinged claw on the end of its long vertical stem moves about if the creature becomes agitated by the information received.

A living sculpture should be interesting (appealing to our senses) not only in appearance but also in behavior. A sculpture should respond appropriately as people interact with it, and different people should provoke different responses from the sculpture. I try to imbue my sculptures with both instinctive behaviors and more complex responses. For example, an aggressive viewer may trigger a defense mechanism within the sculpture, while a passive viewer may experience a more subtle and pleasing reaction. The more time the viewer invests in developing a constructive relationship with the sculpture, the more interesting the response should become. Thus, I try to make pieces that show depth of design as well as depth of behavior. Functionality and aesthetics are tightly linked within the art and create a sense of unity and homogeneity.

Scorpibot and Lady Bug

The Scorpibot (Fig. 1) was my first attempt to integrate a neural network into a sculpture. The sculpture's neural network is implemented in hardware, as wires and transistors, instead of in software as neural networks are often built. The Scorpibot is about 6 feet tall and has a black rubbery skin and a fiberglass shell. The piece uses infrared and touch to receive information about the outside world. The Scorpibot moves around by using a tread system like the ones seen on the military tanks everyone is familiar with. The tank base sprouts a long and fierce-looking vertical stem tipped with a single hinged claw. The claw twitches and moves up and down if the artwork becomes agitated. Considering the simplicity of its brain, this robo-sculpture was able to manifest quite surprising behaviors.

Lady Bug Caption

Fig. 2. The cast aluminum Lady Bug (1993), 20 x 30 x 17 in., rolls on three wheels, listening for an appealing human voice. When in hears a voice pitch it "likes," it will drive over to that person.

The Lady Bug (Fig. 2) is made of cast aluminum. It rolls on 3 wheels and is constantly listening for human voices. If it finds the pitch of a voice appealing the sculpture will drive to the person for a visit. The brain uses a combination of stereoscopic sound locating hardware and a software neural network to control the piece. The neural network listens for voices of a particular pitch, while the hardware tries to locate the source of the sound. Although the brain of this artwork is based in software and can be trained to respond to people's voices, the training process is manual and the sculpture cannot learn new behaviors on its own.

Octofungi

Octofungi (Fig. 3) is an octal-symmetric (eight-legged) polyurethane sculpture about 12 inches tall. Octofungi's reactive dance atop its podium resembles the graceful movements a sea anemone in calm water. Octofungi's nervous system is a neural network that makes decisions based on its eight simple eyes and its history of interactions. Once Octofungi has learned to recognize the environment in which it is located, it relaxes until something happens to change the environment, at which point it will become agitated until it has gotten accostomed to the change. Octofungi learns its environment by adjusting the strengths of its neural connections towards the long term average of what it sees around it. Usually, thirty seconds or so of a nonchanging environment is enough to relax Octofungi. Then, if something changes in the environment, Octofungi will recognize the change and will evaluate its magnitude and impetuousness. The change will trigger either fear or curiosity, and the brain will instruct the legs to move accordingly. Occasionally when nothing much is happening, Octofungi will become bored and examine its surroundings more closely.

Fig. 3. The eight-legged Octofungi (1994-96), a 12-inch-tall sculpture of colored polyurethane, micro glass beads, and natural fibers. Driven by a neural network, Octofungi moves its legs in graceful patterns somewhat resembling the movements of a sea anemone.

Octofungi's legs use shape memory alloy as their motive force. This interesting material is shaped like a fine wire, yet has the unique ability to contract when an electrical current is passed through it. The movement produced by the wire is extremely non-linear, producing a pleasing, life-like motion. Also, the wire is silent, so there is no sound of motors or solenoids to mar the aesthetics of the sculpture.

As originally envisioned, Octofungi would have four processors working in parallel. Two of the processors would act as the forebrain, handling interaction with the outside world, while the other two would act as the hindbrain, handling low-level communication with the rest of the artwork. Currently, Octofungi uses neural networks in two of the processors, while the other two are simple algorithmic feedback controllers which manipulate the legs according to the brains' instructions.

Octofungi learns its environment by adjusting the strengths of its neural connections towards the long-term average of what it sees around it. Usually, thirty seconds or so of a non-changing environment is enough to relax Octofungi. Then, if something changes in the environment, Octofungi will recognize the change, and will evaluate the magnitude and impetuousness of the change. The change will trigger either fear or curiousity, and the brain will instruct the legs to move accordingly.

Gene Genie

Octofungi is still changing. The next version of its brain has yet to be integrated into the artwork. The software used to develop Octofungi's new personality creates a population of brains within a host computer that breeds them for interesting behavior. This 'brain breeding' software, called Gene Genie, incorporates the computer equivalent of time-dependent hormonal flows, hyper-dimensional neural geometry, electrical neural noise, and both DNA and reproductive structuring based on biological models. The software creates a population of virtual neural network brains. The neural structure of the brains is genetically determined, as are the brains' hormonal flow patterns. In addition, the software emulates the electrical 'noise' associated with biological neural structures. The combination of neural networks and time-dependent hormonal flow can potentially create brains that are much more flexible and 'fuzzy' than conventional neural networks. For example, the injection of a particular hormone into the brain can simulate an emotional change, thus changing the brain's behavior. As the hormone slowly spreads throughout the brain, it may trigger other hormone releases, modifying or exacerbating the behavioral change. As this is happening, electrical noise may allow unrelated sections of the brain to communicate weakly with one another, giving the partitions a sort of gestalt of the brain's state.

As the program runs, brains are built on the fly according to their genetic model. They are then tested for interesting behavior, and are allowed to breed with one another depending upon their performance. These processes mimic the biological notion of survival of the fittest and tend to produce populations of brains with a desired type of behavior. The user can specify the desired behavior by way of a fitness function. The program comes with an extensive library of elementary fitness functions, which can be combined together modularly to produce more elaborate objectives. These modules, in turn, can be combined again to produce still more elaborate objectives, ad nauseum.

In addition, the program allows a population of brains to be 'seeded' from a particular brain, allowing one to evolve individual brains for particular traits and then combine the populations together to allow the evolutionary process to determine the best way of merging these behaviors into a complex system. Finally, Gene Genie allows a finished brain to be exported as source code, so that programmers can use the resulant brain in real-world projects.

Octofungi's Paris show

Octofungi's Paris Show

Symbiotic Sculpture

Another idea of interest to me is the concept of creating symbiotic robotic/biological sculptures. The first project I have worked on of this nature was a relationship between a sculpture and an ant colony. However, despite my enthusiasm for the project, I discovered that it is extremely difficult to achieve a truly harmonious and harmless ecosystem. The survival and prosperity of the living organism is extremely important to me. I want it to be able to benefit from the relationship, while the sculpture benefits in turn from the biological component. My main goal for this type of project was to merge two species that would normally kill one another into a sculpture in such a way that all three would interact in a friendly and cooperative way.

My ultimate goal is to create sculptures that can replicate, and consequently, have the ability to change their form and behavior as the generations pass. One idea I have been investigating is creating a cellular sculpture. The basic concept involves a multitude of self-contained cells that can interconnect themselves much like a self-propelled Erector Set. The parent sculptures would need to be sophisticated enough to assembly a copy of themselves and imbue the child with a new genetic code. The parents would need to undergo courtship and sexual reproduction in order to create the child's genetic pattern. After a few generations, the parents would die and would become spare cells for future generations to use. It is obvious that such a project would be extremely costly because of the number of cells necessary for such a cycle to persist, but hopefully, my less ambitious projects will provide the necessary funding.

I am currently working with three engineering assistants: Jason Harris, Michael Hudson, and Keith Causey, who are helping me in the quest for more life-like sculptures. Future projects will incorporate greater autonomy and behavioral intelligence, genetic reproduction, and bodily functions such as eating and digesting. To this end, I am investigating new sources of power such as a distillation process in which the piece would eat organic matter and convert it into usable energy for fuel batteries. Recently planned is a six-foot-high device with three large graceful 'leaves' that unfurl to face in any direction. These 'leaves' duplicate the pattern on the opposite side of the creature in similar fashion to the camouflage of the squid or octopus. It is tentatively dubbed 'Triloterrabyte'.

I have encountered many people who ask me whether robotic sculpture is art, or some misdirected attempt at science. I can't really answer this question; history will determine the nature of robotic sculpture. However, I believe that science is an art, and art is a science, and the two become separated only when society becomes a culture of specialists. Art needs to evolve in order to survive as a living discipline, and can no longer exist as a static canvas without becoming a part of human history. Art must liberate itself and become a living participant in our dreams. As art continues to evolve, classifications between art and science will continue to become more fuzzy and the world will continue to become more interesting.