As a child, you may have fantasized what it would be like to have a robotic friend like R2-D2 from Star Wars. Even now, you probably still have a vivid mental image of the Terminator saying, “I’ll be back.” Real life robots are no less impressive; do you know of any human who can send us pictures from Mars? The Martian rovers do that and much more. So would you consider the Martian rovers to be “alive,” or at least fairly close? If so, then you are in for a surprise. According to scientists Hod Lipson and Jordan B. Pollack at Brandeis University, the closest thing to artificial life is yet a few odd-looking plastic structures crawling awkwardly on the carpet of the laboratory floor.

The question lies in the definition for “life.” The textbook definition for life requires the life form to be made up of cells and be able to carry out the following processes: metabolism; growth; response to stimuli; reproduction and evolutionary adaptation. Several of these requirements, namely cellularity, metabolism, and growth, do not apply well to artificial life, mainly because artificial life tends to be made of inorganic material. However, we can reasonably include the other requirements, namely response to stimuli, reproduction, and evolutionary adaptation, in defining artificial life. Scientists have long been able to create machines that can respond to environmental stimuli. The Martian rovers, for instance, interact with its environment in very complex ways. Yet, until now, scientists have not been able to create a mechanical system to allow robots to self-reproduce or undergo evolutionary adaptation.

Scientists Lipson and Pollack (Lipson and Pollack, 2000) have made that breakthrough. With the help of computer simulation, they have created a system in which robots reproduce themselves without human intervention. Furthermore, these robots can adapt to a physical demand, evolve, and become increasingly fit with each new generation. Although these curious looking, plastic models may look like they jumped right out of some post-modernist abstract painting, they are in this respect closer to being alive than any robot previously created.

The evolution of this robotic species takes place in two phases. In phase one, computer simulation creates virtual robots and makes them evolve according to the principle of natural selection. In phase two, virtual robots are turned into three-dimensional objects. At the end of these two phases, a robot that is fitter than its ancestors will have been created by mechanical means.

In both simulation and reality, these robots have building blocks that are analogous to the body parts of animals. Extendable bars, connected at ball joints, make up the body of the robot and function as muscles and bones. Artificial neurons and synapses form the brain of the robot. The neurons may connect to the bars and make them extend or contract. This movement allows the robot to move around.

Inside the simulator, a fast-forwarded version of evolution takes place. The simulator first creates 200 virtual robots in the simplest form: having zero bars and zero neurons. Then, with each successive generation, the simulator arbitrarily adds, removes or modifies the building blocks of the robots. For example, the simulator may create a bar, change its length or location, or connect it to another bar. It may also alter the weight and connectivity of the neurons. The robots’ simple architecture is greatly advantageous in allowing flexibility in this process. A wide variety of different virtual models is created, each individual having the same fundamental building blocks but variations in structure.

Then the simulator assesses the fitness of each virtual robot. In this model of evolution, fitness is defined as speed, i.e. how far a robot can travel in a given period of time. An individual is considered more fit if it can move faster than its peer. The simulator then modifies the fitter individuals, creating offspring with slightly different morphology, and releasing the offspring back into the population so that their fitness may be assessed in comparison to the other virtual robots.

Usually, several tens of simulated generations pass before the creation of the first moving virtual robot. Then, several hundred generations later, one or more lineages of robots would have attained a considerable level of fitness. At this point, phase two takes place, and the fittest models are molded into physical objects with the help of automatic prototyping technology. As a machine uses thermoplastic material to build up the body of the robot layer by layer, a robot child is being born completely from mechanical means.

            The only things that humans are useful for in this process are attaching a motor to the newborn robot and installing the neurons on a microcontroller to activate the motor. But one would imagine that it would not be too difficult to come up with mechanical ways to perform these actions.

            However, an important characteristic of life is lacking in these robots: sensory. Contrary to other models that may be more complex in structure and programming, the robots of Lipson and Pollack cannot sense or respond to environmental stimuli. As a future research topic, this research team would like to combine sensory and more sophisticated structure with reproductive and evolutionary ability. Imagine the Martian rovers, alone on Mars’s barren terrain, making little rovers all by themselves…

            For now, we can rely on our smart machines to help us perform difficult tasks, even though their complicated functions and programming do not make them “alive.” However, Lipson and Pollack’s research has opened up a huge field of possibilities. One day, all machines may be able to reproduce and improve themselves without human intervention. Humans will no longer have to work to create machines that are better adapted to a certain function; they can just sit back and let evolution take its course. The production of new machines will cost much less, since we no longer need to pay engineers for their creativity.

            Mechanical engineers BEWARE: self-reproducing robots may one day eliminate your jobs!

 

References:

 

Brandeis University. Science Catches up to Science Fiction: Brandeis DEMO Lab moves the world

closer to robot evolution. 2000. Brandeis University News. 

< http://www.brandeis.edu/news/golem.html>. [Accessed 15 Feb 2004]

 

Columbia University Press. 2003.Characteristics of Life. In: The Columbia Electronic

Encyclopedia., 6th ed.

<http://www.infoplease.com/ce6/sci/A0859281.html>. [accessed 5 Mar 2004]

 

Lipson, Hod and Jordan B. Pollack. 2000. Automatic Design and Manufacture of Robotic Lifeforms.

Nature. 406: 974-978.

 

Lipson, Hod and Jordan B. Pollack. 2000.The Golem Project.

< http://helen.cs-i.brandeis.edu/golem/>. [accessed 15 Feb 2004]