TY - JOUR
T1 - Toward energy Autonomy in heterogeneous Modular Plant-Inspired Robots through Artificial evolution
AU - Veenstra, Frank
AU - Metayer, Chloe
AU - Risi, Sebastian
AU - Stoy, Kasper
PY - 2017
Y1 - 2017
N2 - Contemporary robots perform energy intensive tasks—e.g., manipulation and locomotion—making the development of energy autonomous robots challenging. Since plants are primary energy producers in natural ecosystems, we took plants as a source of inspiration for designing our robotics platform. This led us to investigate energy autonomy in robots through employing solar panels. As plants move slowly compared to other large terrestrial organisms, it is expected that plant-inspired robots can enable robotic applications, such as long-term monitoring and exploration, where energy consumption could be minimized. Since it is difficult to manually design robotic systems that adhere to full energy autonomy, we utilize evolutionary algorithms to automate the design and evaluation of energy harvesting robots. We demonstrate how artificial evolution can lead to the design and control of a modular plant-like robot. Robotic phenotypes were acquired through implementing an evolutionary algorithm, a generative encoding and modular building blocks in a simulation environment. The generative encoding is based on a context sensitive Lindenmayer-System (L-System) and the evolutionary algorithm is used to optimize compositions of heterogeneous modular building blocks in the simulation environment. Phenotypes that evolved from the simulation environment are in turn transferred to a physical robot platform. The robotics platform consists of five different types of modules: (1) a base module, (2) a cube module, (3) servo modules, and (4,5) two types of solar panel modules that are used to harvest energy. The control system for the platform is initially evolved in the simulation environment and afterward transferred to an actual physical robot. A few experiments were done showing the relationship between energy cost and the amount of light tracking that evolved in the simulation. The reconfigurable modular robots are eventually used to harvest light with the possibility to be reconfigured based on the needs of the designer, the type of usable modules, and/or the optimal configuration derived from the simulation environment. Long-term energy autonomy has not been tested in this robotics platform. However, we think our robotics platform can serve as a stepping stone toward full energy autonomy in modular robots.
AB - Contemporary robots perform energy intensive tasks—e.g., manipulation and locomotion—making the development of energy autonomous robots challenging. Since plants are primary energy producers in natural ecosystems, we took plants as a source of inspiration for designing our robotics platform. This led us to investigate energy autonomy in robots through employing solar panels. As plants move slowly compared to other large terrestrial organisms, it is expected that plant-inspired robots can enable robotic applications, such as long-term monitoring and exploration, where energy consumption could be minimized. Since it is difficult to manually design robotic systems that adhere to full energy autonomy, we utilize evolutionary algorithms to automate the design and evaluation of energy harvesting robots. We demonstrate how artificial evolution can lead to the design and control of a modular plant-like robot. Robotic phenotypes were acquired through implementing an evolutionary algorithm, a generative encoding and modular building blocks in a simulation environment. The generative encoding is based on a context sensitive Lindenmayer-System (L-System) and the evolutionary algorithm is used to optimize compositions of heterogeneous modular building blocks in the simulation environment. Phenotypes that evolved from the simulation environment are in turn transferred to a physical robot platform. The robotics platform consists of five different types of modules: (1) a base module, (2) a cube module, (3) servo modules, and (4,5) two types of solar panel modules that are used to harvest energy. The control system for the platform is initially evolved in the simulation environment and afterward transferred to an actual physical robot. A few experiments were done showing the relationship between energy cost and the amount of light tracking that evolved in the simulation. The reconfigurable modular robots are eventually used to harvest light with the possibility to be reconfigured based on the needs of the designer, the type of usable modules, and/or the optimal configuration derived from the simulation environment. Long-term energy autonomy has not been tested in this robotics platform. However, we think our robotics platform can serve as a stepping stone toward full energy autonomy in modular robots.
KW - modular robots
KW - plant-inspired robots
KW - artificial intelligence
KW - energy autonomy
KW - evolutionary algorithms
KW - modular robots
KW - plant-inspired robots
KW - artificial intelligence
KW - energy autonomy
KW - evolutionary algorithms
U2 - 10.3389/frobt.2017.00043
DO - 10.3389/frobt.2017.00043
M3 - Journal article
SN - 2296-9144
VL - 4
JO - Frontiers in Robotics and AI
JF - Frontiers in Robotics and AI
M1 - 43
ER -