The development of a robust symbolic communication system for robots via embodied iterated imitation

Abstract

In this research, real robots are used to model the emergence and development of a symbolic communication system in a group of social agents. It is shown that as the predefined symbols of the language are transmitted among the robots, variations due to embodiment arise in the learned symbols. As a result of these variations, transmissibility of the randomly generated symbols of the language decreases. We introduce multiple cycles of iterated imitation during which the symbols of the language evolve and adapt to the uncertainties in robots' sensors and actuators. Specific structures emerge in the learned symbols and transmissibility of the language increases. This process is proposed as a model that explains how structural syntax emerges and evolves in a symbolic communication system.

Evolution of combinatorial structure in learned forms through embodied iterated learning in a robot collective

“…To observe and replicate the movement trajectories of others, the robots use an embodied robot-to-robot imitation algorithm (Erbas, 2019) that purely depends on the on-board sensors of the robots. The algorithm works as follows:…”

“…In this way, robots are able to observe and replicate the action sequences of other robots. Further details about the embodied robot-to-robot movement imitation algorithm can be seen in Erbas (2019). As stated above, the movement imitation algorithm purely depends on the sensors of the robots.…”

“…It was shown that initial learned categorization of body position/motor command couples was sufficient for the learning of a shared lexicon through interaction. In another article, Erbas (2019) presented the concept of imitation games in which mobile robots interacted with each other to guess the meaning of the behaviors that constitute a predefined lexicon. It was stated that the transmissibility of the learned behaviors increased as a result of multiple cycles of learning and use.…”

Modeling Active Learning in a Robot Collective

“…The learner robot can learn a demonstrated movement pattern by using a movement imitation algorithm [18]. The algorithm works as follows:…”

“…In this way, the learner robot is able to observe and learn the movement patterns that are demonstrated by the teacher robot. Further details about the movement imitation algorithm and an analysis on the type of copying errors can be seen in [18].…”