image: An assembly of identical robotic units that each try to adapt their own behavior to achieve locomotion. view more
Credit: Image credit: Soft Robotic Matter Group.
Researchers report a modular robot that autonomously adapts to its environment to achieve optimal behavior. Imbuing robots with the ability to detect their environments and to adapt and behave autonomously under a range of conditions remains a challenge. Previous approaches have increased the complexity of the central controller or implemented machine learning. Johannes T.B. Overvelde and colleagues developed and tested a modular robot, assembled from identical discrete units that individually and continuously sense and respond to their environment. The authors set the modular robot, programmed with a Monte Carlo scheme, on a circular track to see how it developed locomotion. The robot learned to move forward and achieved maximum speed after 80 seconds. In this and other experiments, the modules were able to maintain optimal behavior, even when sustaining damage, as long as their memory was made to represent the current environment. No communication between the modules except for a physical connection was needed, eliminating the need for a central controller. According to the authors, scaling up such an approach could result in robotic materials that could be miniaturized and autonomously learn to navigate environments for a range of applications in healthcare, disaster relief, and space exploration.
Article #20-17015: "Continuous learning of emergent behavior in robotic matter," by Giorgio Oliveri, Lucas C. van Laake, Cesare Carissimo, Clara Miette, and Johannes T.B. Overvelde.
MEDIA CONTACT: Johannes T.B. Overvelde, AMOLF, Amsterdam, NETHERLANDS; email: overvelde@amolf.nl
###
Journal
Proceedings of the National Academy of Sciences