Embodied evolving systems — biological or robotic — are multi-hierarchical networks of structure, function, information, and behavior. What general principles operate and govern these systems? What types and levels of explanations are sufficient? What kinds of information, theory, and model allow us to predict and, ultimately, control embodied evolving robots? Can insights from evolutionary robotics illuminate principles underlying biological evolving systems, or general analytical frameworks for studying embodied evolving systems?
These are ambitious questions. To make significant headway in answering them is the goal of this research topic, with the burgeoning field of evolutionary robotics as our methodological and theoretical center of gravity.
In evolutionary robotics, a population of agents is autonomously designed by feedback between the agents and the environment in which and with which they interact. Fitness functions, created a priori by investigators, represent the task, behavior, or goal to be evolved. Random processes, including mutation and reproduction, create variation and allow a population to explore novel combinations of traits in a search for local and global optima. As environments change, highly evolvable populations respond quickly to the changes.
But as causal mechanisms, selection and random processes alone may be insufficient to model and understand the full range of evolutionary possibilities. What else matters? Biology indicates: (1) the genome encodes and decodes information as physical machinery in the context of its physical environment; (2) development processes require and respond to feedback from the growing agent and its environment; (3) plasticity from learning and adjustable morphology allow for non-genetic responses that buffer environmental variation, and (4) dynamics of ecological interactions point to the population as both part of the environment and a super-agent.
Evolvability, environments, embodiment and emergence are important concepts related to the evolution of robots and robotic systems. We encourage participants in this research topic to focus on intersections and interactions of these features in experiments, simulations, models, theoretical analyses, and reviews. We encourage collaboration, elaboration, argument, and disagreement.
We open this research topic with the following questions.
If evolvability is the capacity of a population of agents for future change, then what features of the current population are the best predictors? What types of mechanisms or properties — e.g., selection, mutation, duplication, mixability, modularity, genotype-phenotype mapping — enhance evolvability?
If robustness of an agent or system is measured by its ability to produce the same behavior in different environments, then how can robust robotic systems evolve? What costs does robustness incur? What information do we need to accurately predict robustness?
If embodiment recognizes that behavior is a dynamic process not controlled solely by brain or controller, then what consitutes physical “information” and how is it manipulated by the agent, system, or environment? Can we use reductionism to study embodiment, given that embodiment requires explanation at the level of the whole agent and its environment?
If emergence is the appearance of a property in a system that can’t be explained by a study of its components in isolation, then what causal role, if any, does it play in our understanding of behavior and evolution? How does one study emergence scientifically? What are the various concepts subsumed by the term “emergence,” and which of those concepts, if any, are useful for explaining how evolving systems work?
Embodied evolving systems — biological or robotic — are multi-hierarchical networks of structure, function, information, and behavior. What general principles operate and govern these systems? What types and levels of explanations are sufficient? What kinds of information, theory, and model allow us to predict and, ultimately, control embodied evolving robots? Can insights from evolutionary robotics illuminate principles underlying biological evolving systems, or general analytical frameworks for studying embodied evolving systems?
These are ambitious questions. To make significant headway in answering them is the goal of this research topic, with the burgeoning field of evolutionary robotics as our methodological and theoretical center of gravity.
In evolutionary robotics, a population of agents is autonomously designed by feedback between the agents and the environment in which and with which they interact. Fitness functions, created a priori by investigators, represent the task, behavior, or goal to be evolved. Random processes, including mutation and reproduction, create variation and allow a population to explore novel combinations of traits in a search for local and global optima. As environments change, highly evolvable populations respond quickly to the changes.
But as causal mechanisms, selection and random processes alone may be insufficient to model and understand the full range of evolutionary possibilities. What else matters? Biology indicates: (1) the genome encodes and decodes information as physical machinery in the context of its physical environment; (2) development processes require and respond to feedback from the growing agent and its environment; (3) plasticity from learning and adjustable morphology allow for non-genetic responses that buffer environmental variation, and (4) dynamics of ecological interactions point to the population as both part of the environment and a super-agent.
Evolvability, environments, embodiment and emergence are important concepts related to the evolution of robots and robotic systems. We encourage participants in this research topic to focus on intersections and interactions of these features in experiments, simulations, models, theoretical analyses, and reviews. We encourage collaboration, elaboration, argument, and disagreement.
We open this research topic with the following questions.
If evolvability is the capacity of a population of agents for future change, then what features of the current population are the best predictors? What types of mechanisms or properties — e.g., selection, mutation, duplication, mixability, modularity, genotype-phenotype mapping — enhance evolvability?
If robustness of an agent or system is measured by its ability to produce the same behavior in different environments, then how can robust robotic systems evolve? What costs does robustness incur? What information do we need to accurately predict robustness?
If embodiment recognizes that behavior is a dynamic process not controlled solely by brain or controller, then what consitutes physical “information” and how is it manipulated by the agent, system, or environment? Can we use reductionism to study embodiment, given that embodiment requires explanation at the level of the whole agent and its environment?
If emergence is the appearance of a property in a system that can’t be explained by a study of its components in isolation, then what causal role, if any, does it play in our understanding of behavior and evolution? How does one study emergence scientifically? What are the various concepts subsumed by the term “emergence,” and which of those concepts, if any, are useful for explaining how evolving systems work?
