Elsevier

Biosystems

Volume 107, Issue 1, January 2012, Pages 34-51
Biosystems

Behavioral robustness: An emergent phenomenon by means of distributed mechanisms and neurodynamic determinacy

https://doi.org/10.1016/j.biosystems.2011.09.002Get rights and content

Abstract

Theoretical discussions and computational models of bio-inspired embodied and situated agents are introduced in this article capturing in simplified form the dynamical essence of robust, yet adaptive behavior. This article analyzes the general problem of how the dynamical coupling between internal control (brain), body and environment is used in the generation of specific behaviors. Based on the Evolutionary Robotics (ER) paradigm, four computational models are described to support discussions including descriptions on performance after a series of structural, sensorimotor or mutational perturbations, or are developed in the absence of them. Experimental results suggest that ‘dynamic determinacy’ – i.e. the continuous presence of a unique dynamical attractor that must be chased during functional behaviors – is a common dynamic phenomenon in the analyzed robust and adaptive agents. These agents show dynamical states that are definitely and unequivocally characterized via transient dynamics toward a unique, yet moving attractor at neural level for coherent actions. This determinacy emerges as a control strategy rooted on behavioral couplings and relies on mechanisms that are distributed on brain, body and environment. Different ways to induce further distribution of behavioral mechanisms are also discussed in this paper from a bio-inspired ER perspective.

Introduction

What ‘ensures’ biological robustness? This is one of the unanswered questions facing scientists since von Neumann (1956) noted the complexity of such a problem by opening debates on ‘the synthesis of reliable organisms from unreliable components’. Reliability usually refers to robustness that refers to a systemic property commonly attributed to living organisms (Stelling et al., 2004). Despite the lack of a formal definition, robustness stands for the ability of artificial or biological organisms to maintain their capacities (functionalities) in a normal situation and under unexpected internal/external perturbations.

Studies in neuroscience and systems biology (Alon, 2006) generally propose an organism-centred standpoint of robustness. However, the partition between organism and environment is not always helpful for thinking on them as ‘highly interdependent’. This is because studying the internal milieu of organisms only focuses on one-third of the relationship between control systems, body and environment, giving special emphasis on the former component (e.g. the brain or the nervous system). Internal structures as modularity, decoupling and redundancy are conventionally thought to be necessary for robustness in systems biology (Krakauer, 2005). Structures like these may be required to support some systemic functionalities to certain perturbations, but they do not ‘ensure’ robust traits in themselves see Kitano, 2004, Kitano, 2007 and Krakauer (2005) for complementary discussions. An example of this last point comes from the computational paradigm of neural networks used to explore how modularity can lead to more efficient task management (Calabretta et al., 1998). Despite recognized robust properties to noisy data, most modular neural networks can show certain decay in their performance because a considerably high amount of noise can reduce drastically their filtering capacity (Arbib, 1995). Therefore, can we certainly think on robustness as internally generated in small neural-based systems? Answers to what is required for robustness at behavioral level (behavioral robustness) could guide better scientific descriptions of habits, coherent experience and adaptation to changing environments. Answers to that question could represent small steps toward, for example, the understanding of how the brain of simple organisms structures its internal dynamics underpinning most movements.

This paper describes selected studies developed by the author in the Evolutionary Robotics (ER) context (Harvey, 1992), a methodology from computational neurosciences used here to illustrate how the process of natural selection can lead to the evolution of robust and adaptive traits. In particular this work proposes experiments where agents cannot exclusively relay on internal control for robustness. As we will see later in this article, agents can still behave coherently despite certain levels and types of perturbations by exploiting systemic features like situatedness, embodiment, and agent–environment coupled dynamics. Situatedness, or being situated in the environment, means that agents (biological or artificial organisms) use their surroundings to directly influence their future actions (Brooks, 1991). Embodiment refers to the physical existence of an organism or robot having a co-related, but essentially different, dynamics from the environment (see Ziemke, 2003 for a classification of embodiment). The concept of coupled dynamics will refer forefront to the active interaction between the neurocontroller (‘brain’), body, and environment systems.

The ER methodology is a relatively assumption-free paradigm compared to other synthetic approaches rooted on artificial evolution (Nolfi and Floreano, 2000). Unfortunately, works in ER have so far paid relatively little attention to topics such as distributed mechanisms and cognition (Ziemke et al., 2004) and behavioral robustness (Silverman and Ikegami, 2010). The idea of mechanism that are distributed refers in this article to functional dependencies of internal-control, body and environment dynamics that a control system shows at agent–environment level.

By limiting experimental analyses to some case studies from computational modeling, this paper highlights behavioral robustness as a dynamical process (rather than a structural property), being in any case certainly incomplete if we do not focus on engaged brain–body–environment dynamics. In other words, the described studies can show us that the analyzed robustness is better understood in the context of dynamical couplings, not in terms of internal (neural) assemblies. These dynamical couplings, however, are not always the full determinants of robustness (Fernandez-Leon, 2010a). The reported models verify the power of the ER methodology to explore dynamical mechanisms for behavioral robustness in artificial agents.

The overall goal of this paper is to motivate the reader to see distributed functionality as beneficial for behaviorally robust systems, instead of robustness in neural systems being entirely determined ‘from inside’. This work addresses current discussions on distributed behavioral mechanisms as central to the emergence of cognitive processes and robust behaviors in ER. Dynamical Systems Theory (Strogatz, 1994) provides the conceptual framework for the synthesis and analysis of autonomous agents and their environments.

Sections 2 A perspective shift of behavioral robustness, 3 Artificial evolution as a tool for robustness research discuss the approach presented in this paper and a short survey is given on what has been reported in associated literature. Section 4 describes four experiments in the ER field highlighting the common dynamical features between these models. Finally, the main experimental lessons, dynamical observations, and further discussions are identified for future studies from Sections 5 Experimental lessons, 6 Transient dynamics and dynamic determinacy, 7 Have we evidenced biologically conceivable robustness?, 8 Discussions.

Section snippets

A perspective shift of behavioral robustness

The accepted understanding of what produces robust and adaptive behavior is gradually changing from being generated by isolated control mechanisms within organisms toward dynamical processes occurring over multiple and distributed systemic components (Calcott, 2010, Kitano, 2002). However, the word ‘distributed’ in neuroscience and Artificial Intelligence (AI), for instance, still means distributed within the brain like distributed parallel computation in neural networks (Fernandez-Leon, 2010a

Artificial evolution as a tool for robustness research

In the development of simulated organisms (agents) by means of the Darwinian selection, ER has increased in popularity across the computational neuroscience and robotics research fields. ER is a method to automatically generate control systems that are comparatively simpler or more efficient than those engineered with qualitatively similar design techniques (Floreano and Keller, 2010). An ER agent can be defined with a simple internal structure with no special mechanisms for ensuring adaptive

A sensor-based goal-seeking task under neural perturbations

Biological robustness is generally discussed in systems biology literature as a by-product of evolution, where robust mechanisms emerge from noisy processes (Félix and Wagner, 2008). In a neural context, the incidence of internally generated neural noise on minimal situated, embodied and dynamical agents is investigated so far in Fernandez-Leon and Di Paolo, 2007, Fernandez-Leon and Di Paolo, 2008 and Fernandez-Leon et al. (2009) (see also Jakobi, 1998). These articles inspire some of the ideas

Experimental lessons

Taking the dynamics of the environment (including body) into account generally makes the study of robustness a hard problem, even more difficult. The reported experiments in this article consequently have concentrated on minimal models and analyzed dynamically contributions of agent–environment to robust and adaptive behavior. The general experimental observations are listed as follows:

  • Lesson 1: Not all distribution of agentsbehavioral mechanisms are equally helpful in generating robustness.

Transient dynamics and dynamic determinacy

The described experiments suggest us a common dynamical phenomenon associated to the type of behavioral robustness explained in this article: ‘dynamical determinacy’, i.e. the continuous presence of a unique dynamical attractor that must be chased during behavior (Fig. 13-left). For instance in Section 4.2, the dynamical trajectories toward one of the two-fixed points at neurocontroller level chiefly determine the long-term avoiding behavior of the studied bistable agent. This happens when the

Have we evidenced biologically conceivable robustness?

A novel aspect of the present paper may be found in how the scheme of coupled, controller–body–environment dynamics could come to terms with the continuous presence of a unique dynamical attractor that must be chased during behaviors. The overall observation is that by introducing as few assumptions as possible about the nature of robust behaviors, experiments in this paper show us a common dynamical control through models rooted on the emergence of a unique attractor for coherent behavior.

Discussions

The key assumption used in this article is that biological organisms have evolved in coupled, controller–body–environment conditions. The briefly described models showing such a coupling allow systematic tests that are not currently amenable to experimental techniques in the biological realm. The method proposed here (further developed in Fernandez-Leon, 2010a and associated publications) is taken as a starting point to further develop the introduced approach. However, the experimental models

Acknowledgements

The author would like to thank to Dr. Ezequiel Di Paolo and Dr. Tom Froese for discussions on an early version of part of this work; Dr. Andy Philippides and Dr. Takashi Ikegami for suggestions and helpful comments for related work to this article. Also, thanks to members of CCNR (University of Sussex) and anonymous reviewers for useful comments in how to improve discussions within this manuscript.

References (51)

  • U. Alon

    An Introduction to Systems Biology: Design Principles of Biological Circuits

    (2006)
  • R. Beer

    On the dynamics of small continuous-time recurrent neural networks

    Adapt. Behav.

    (1995)
  • R. Beer

    The dynamics of active categorical perception in an evolved model agent

    Adapt. Behav.

    (2003)
  • R. Beer

    Autopoiesis and cognition in the game of Life

    Artif. Life

    (2004)
  • R. Beer

    Parameter space structure of continuous-time recurrent neural networks

    Neural Comput.

    (2006)
  • Beer, R. Dynamical analysis of evolved agents: a premier. In: Vargas P., Di Paolo E., Harvey I., Husbands P., The...
  • R. Brooks

    New approaches to robotics

    Science

    (1991)
  • C. Buckley et al.

    Monostable controllers for adaptive behaviour

  • R. Calabretta et al.

    Emergence of functional modularity in robots

  • B. Calcott

    Wimsatt and the robustness family: review of Wimsatt's re-engineering philosophy for limited beings

    J. Biol. Philos.: Rev. Essay

    (2010)
  • A. Calvitti et al.

    Analysis of a distributed model of leg coordination. I. Individual coordination mechanisms

    Biol. Cybern.

    (2000)
  • M. Félix et al.

    Robustness and evolution: concepts, insights and challenges from a developmental model system

    Heredity

    (2008)
  • Fernandez-Leon, J.A. (2010a). Behavioural robustness and the distributed mechanisms hypothesis. DPhil Thesis....
  • J.A. Fernandez-Leon

    Evolving experience-dependent robust behaviour in embodied agents

    Biosystems

    (2010)
  • Cited by (5)

    • How simple autonomous decisions evolve into robust behaviours?: A review from neurorobotics, cognitive, self-organized and artificial immune systems fields

      2014, BioSystems
      Citation Excerpt :

      More specifically, this work mainly focuses on the dynamical coupling between an organism’s modeled central nervous system (in terms of cybernetics, the robot’s control systems) within bodies (the robot’s mechanical support) and the environments under the effects of internal and external factors (perturbations) disturbing the organism’s behaviors. The aim of this work is consequently to understand behaviors that greatly outlast any initiating stimulus by means of processes that arise from organism–environment interactional level (Fernandez-Leon, 2012). The purpose of this understanding is to examine how desired behaviors emerge and how we can control them in order to give light through synthetic experiments to build more robust and bio-inspired systems and robots.

    View full text