© 2004 Robert A. Freitas Jr. and Ralph C. Merkle. All Rights Reserved.

Robert A. Freitas Jr., Ralph C. Merkle, Kinematic Self-Replicating Machines, Landes Bioscience, Georgetown, TX, 2004.

2.3.7 Embodied Evolution: Algorithmic Replication

A more limited form of self-replication, originally addressed by Husbands and Harvey [577] and others [578-593] as “evolutionary robotics,” has recently been investigated experimentally as “embodied evolution” by Watson et al [594]. In these experiments, an evolutionary algorithm is distributed amongst and embodied within a population of eight physical mobile robots that “reproduce” with one another and “evolve” together, in a specialized task environment (e.g., a demonstration of phototaxis). Although the robot population is fixed in size and membership, operating algorithms are transferred among the various robot members by “broadcasting a gene,” analogous to the sharing of genetic information by promiscuous microorganisms. There is no “reproduction mode” as such because reproduction is concurrent with task behavior [582]. According to the researchers [594]: “Assuming that we cannot really create new robots spontaneously, the offspring must be implemented using (other) robots of the same population. And, if the robots do not have structurally reconfigurable bodies, reproduction must simply mean the exchange of control program code.” This approach “enables the study of the effects of integrating reproduction with other autonomous behaviors into real robots in a manner that has previously only been possible in simulated ALife experiments,” although the authors caution that “reproduction may interfere with task behavior.”

The authors [594] point out that the artificial life literature provides several examples [442, 595-597] of simulated systems where agent behavior and reproductive activity are integrated, but heretofore experiments using physical robots have not been able to integrate reproduction with other autonomous behaviors. Also noteworthy are the attempts by Lund et al [600] to evolve (in simulation) both a robot control program and some parameters of the robot’s physical body including number of sensors, sensor positions, body size, turret radius, and so forth. The evolution can explore only those parameters foreseen by the designer and cannot produce “creative” new designs, consistent with our desire for our machines to exhibit “safe” replication (Section 5.11).

Last updated on 1 August 2005