Resilient Self-Reproducing Systems

This paper is motivated by the need to minimize the payload mass required to establish an extraterrestrial robotic colony. One approach for this minimization is to deploy a colony consisting of individual robots capable of self-reproducing. An important consideration once such a colony is established is its resiliency to large-scale environment or state variations. Previous approaches to learning and adaptation in self-reconfigurable robots have utilized reinforcement learning, cellular automata, and distributed control schemes to achieve robust handling of failure modes at the modular level. This work considers self-reconfigurability at the system level, where each constituent robot is endowed with a self-reproductive capacity. Rather than focus on individual dynamics, the hypothesis is that resiliency in a collective may be achieved if: 1) individual robots are free to explore all options in their decision space, including self-reproduction, and 2) they dwell preferentially on the most favorable options. Through simulations, we demonstrate that a colony operating in accordance with this hypothesis is able to adapt to changes in the external environment, respond rapidly to applied disturbances and disruptions to the internal system states, and operate in the presence of uncertainty.