Half-day Workshop on
Developmental Plasticity and Evolutionary Robotics
Friday, May 19, 9:00AM-12:20PM,
Location: Room Parallel 1 (Cook), William A. Egan Civic & Convention Center, 555 W. Fifth Avenue, Anchorage, AK 99501
|Angel P. del Pobil||Fumiya Iida|
|Department of Engineering and Computer Science||Department of Engineering|
|Robotics Intelligence Laboratory||Bio-Inspired Robotics Laboratory|
|Jaume-I University, Castellón, Spain||University of Cambridge, Cambridge, UK|
This workshop will address the impact that developmental plasticity can have on evolutionary robotics from a multidisciplinary perspective. Given its fundamental role in driving diversification and speciation, this emerging and growing area of research can possibly contribute to a new paradigm in developmental and evolutionary robotics.
Developmental (phenotypic) plasticity is the capacity of a genotype to express a range of phenotypes in response to distinct environmental conditions. It can be visualized by the use of reaction norms, which plot values for a specific phenotypic trait across two or more environments. Plasticity is an integral part of organismal development that allows for robustness in the face of nearly ubiquitous environmental perturbations, allowing organisms to respond to environmental variation, often in an adaptive manner. The study of phenotypic plasticity has progressed significantly over the past few decades, fostered by the study of its quantitative genetic underpinning, becoming a primary target of investigations in evolutionary biology today, and many plasticity-specific concepts have become well established as part of the mainstream thinking in the field. Much research now focuses on identifying genes and pathways that change expression in the face of environmental changes, generating viable, functional and novel variants.
Environmentally induced changes to phenotypes include well-known discrete morphological polyphenisms (with enormous differences in some insects), but organisms can also alter their biochemistry, physiology, behavior, and life history in response to the environment. In general, behavioral and physiological traits are rapidly reversible within individuals, whereas morphology and life history tend to be permanent. A great many traits fall somewhere in-between.
Given the indisputable fact that natural selection selects not among genotypes, but among phenotypes, the phenotype, and variation among phenotypes, must play a major role in evolution to the extent that evolutionary biologists are increasingly considering that plasticity influences the direction of evolution by allowing survival in novel environments, by facilitating the accumulation and release of cryptic genetic variation -as a counter-balance to mutation-driven evolution- and by producing organic novelty that can then be incorporated into the genome via genetic assimilation. In this encompassing paradigm, the environment not only plays the role of selecting among genetically produced variation, but it also creates phenotypic variation and selects among that variation. Thus, environmentally induced phenotypic variation is becoming a dominant position in evolutionary theory.
Recently, the importance of plasticity has been stressed in the evolution of complex and integrated phenotypes. Since organisms are complex networks of interacting systems, complex plasticities may result from the integration of numerous underlying plastic traits giving rise to a functioning individual of high fitness. Consequently, the focus is moving from the simpler and static view of one trait at one developmental time point, to whole organism and whole-life cycle perspectives.
Evolutionary robotics is a subfield of robotics that pursues the design of more robust and adaptive robots by applying principles of natural evolution such as selection, variation, and heredity. In particular, evolutionary robotics aims to evolve the controllers, the morphologies, or both. Conversely, it can offer insights to evolution theory by means of experiments with evolving robots which stand in for biological organisms. As in embodied intelligence, the right approach should be holistic in the sense that it considers the robot as a whole, including its morphology, sensory apparatus, motor system, control architecture and materials. All of these components interact and jointly, with the environment, determine the robot’s behavior.
In the current state of the art, after two decades of history, artificial neural networks are the preeminent controller paradigm in the field, because they impose little constraints and allow to use raw inputs from sensors and send low-level commands to actuators. Also, simulation is the tool of choice since it makes possible to quickly conduct experiments and evaluate models; however, when the best evolved solutions in simulation are transfered to the hardware robot, most often they do not work. This problem is called the reality gap, a mismatch between simulated and real-world behavior. A more recent trend is to abandon simulators altogether and move towards the emerging area of artificial evolution in physical systems, the so-called evolution of things, opening up new avenues towards autonomous machines that can adapt to their environment. For a maximum match with biological evolution, these systems should be driven by environmental selection only, instead of a user-defined fitness function. Environment-driven evolution has only recently emerged as a paradigm in evolutionary robotics, though it is well known in the field of artificial life.
Indeed, recent results suggest that the classic objective-based fitness function may hinder evolvability, which is one of the main open questions in evolutionary robotics: how to design artificial systems that are as evolvable as natural species. This is also a central open question in evolutionary biology: what makes natural organisms so evolvable, that is, how species quickly evolve responses to new evolutionary challenges. Importantly, developmental plasticity can increase the ‘evolvability’ of a systems in three crucial ways, thereby increasing a lineage’s potential for diversification and innovation: by providing new targets for evolutionary processes, by promoting novelty through ‘re-usable’ building blocks for development, and by creating novel trait interactions [Moczek et al. 2011].
In spite of its many successes, current algorithms oversimplify a number of aspects of biologycal evolution, possibly missing critical features. Conspicuosly, most current evolutionary algorithms represent candidate solutions by “genotypes” which are codes intended to be translated into phenotypes for their fitness to be evaluated. This genotype–phenotype mapping is typically a simple mathematical transformation or parameterized procedure, in contrast with the highly complex biochemical and developmental process influenced by the environment that is found in natural evolution, in which phenotypic plasticity plays a major role, as discussed above. In this regard, a new research trend is generative and developmental representations with the use of indirect encodings and allowing the reuse of code, which helps to scale up the complexity of artificially evolved phenotypes, for instance, also in artificial life and morphogenetic engineering. Only very recently proposals have been put forward in which morphological plasticity is achieved through learning during the lifetime of a virtual creature, allowing it to quickly adapt its morphology to the environment [Krcah 2016]; or a model of artificial genotype is suggested to shape the relationship with the phenotype and the environment in an artificial system, presenting a similar behavior to that of organisms in what regards the concept of norm of reaction [Duran & del Pobil 2016]; or the morphological evolution of physical robots through model-free phenotype development, generating morphological and behavioral diversity [Brodbeck, Hauser & Iida 2015].
Evolutionary robotics claims to be distinct from other fields of engineering in that it is inherently based on a biological mechanism. If faithful to this principle, it should incorporate phenotypic plasticity as one of its major components, given its fundamental role in driving diversification and speciation, possibly contributing to a new paradigm in developmental and evolutionary robotics.
The purpose of this workshop is to address the above issues, in particular how developmental plasticity can play a major role in emerging paradigms in evolutionary robotics. Recent progress in evolutionary biology suggests that the interplay between robotics research and the current understanding of the plasticity mechanisms underlying the development of living organisms is a very promising track to be followed. Interaction between the two fields is useful for both evolutionary robotics, which can take inspiration from biological solutions to engineering problems, and evolutionary biology, that can benefit from artificial emulation of biological mechanisms which can prove the validity of research hypothesis.
We emphasize a highly multi-disciplinary approach to these topics, involving participants from biology, engineering, robotics, and other disciplines. By encouraging integration of multiple perspectives, we aspire to arrive at new insights. The workshop will be organized in such a way as to generate fruitful discussions, it will consist of invited presentations (30 min. each) and regular presentations (15 min. each) with a significant amount of additional time for a panel discussion. A call will be announced to solicit contributions and a website will be set up.The organizers will put together the proceedings of the workshop consisting of accepted and invited contributions in final form, the conclusions resulting from the discussions and other relevant materials.
Please send your abstracts to <email@example.com>, as an attached PDF file. Please include [IJCNN_WS] in the subject.
April 2nd : Extended Abstract Submission Deadline.
April 10th, 2017: Notification of Acceptance.
March 12th : Abstract Submission Deadline.
March 20th, 2017: Notification of Acceptance.
Abstract Format: 1 to 2 pages, plus references, in two-column regular IJCNN paper format. Accepted contributions will be presented at the workshop (15-min oral presentations).
The workshop will be organized in such a way as to generate fruitful discussions, it will consist of invited presentations (30 min. each) and regular presentations (15 min. each) with a significant amount of additional time for a panel discussion.
Proceedings will consist of accepted and invited contributions in final form, the conclusions resulting from the discussions, along with other relevant materials.
We look forward to your active participation
9:00 - 9:10 Welcome and Introduction -- Angel P. del Pobil
9:10 - 9:40 An overview of mechanisms of plasticity in biology -- Emilie Snell-Rood, University of Minnesota, St Paul
9:40 - 10:00 Generative Morphologies and Control for Evolving Heterogeneous Modular Robots -- Frank Veenstra and Kasper Stoy, IT University of Copenhagen, Denmark
10:00 - 10:20 How Morphological Development Can Guide the Evolution of Soft Robots -- Sam Kriegman and Josh Bongard, University of Vermont, Burlington
10:20 - 10:40 Phenotypic plasticity in a robotic system based on a model of artificial genotype -- Angel J. Durán and Angel P. del Pobil, Jaume-I University, Spain
10:40 - 11:00 Break
11:00 - 11:30 Inexpensive, Distributed, Open Source Evolutionary Robotics -- Josh Auerbach, Champlain College, Burlington, Vermont
11:30 - 11:50 Adaptation of Virtual Creatures to Different Environments Through Morphological Plasticity -- Peter Krcah, Charles University, Prague, Czech Republic
11:50 - 12:20 Discussion and conclusion
A. J. Duran and A. P. del Pobil, A Model of Artificial Genotype and Norm of Reaction in a Robotic System, E. Tuci et al. (Eds.): SAB 2016, LNAI 9825, pp. 267–279, 2016.
Grzyb, B.J., Smith, L.B., del Pobil, A.P., 2013, Reaching for the unreachable: reorganization of reaching with walking, IEEE Transactions on Autonomous Mental Development, Vol. 5, No. 2, pp. 162-172.
Chinellato, E., Antonelli, M., Grzyb, B.J., del Pobil, A.P., 2011, Implicit Sensorimotor Mapping of the Peripersonal Space by Gazing and Reaching, IEEE Transactions on Autonomous Mental Development, Vol. 3, No. 1, pp. 43-53.
Pfeifer R, Lungarella M, Iida F (2007), Self-organization, embodiment, and biologically inspired robotics. Science 318: 1088–1093.
Brodbeck L, Hauser S, Iida F (2015), Morphological Evolution of Physical Robots through Model-Free Phenotype Development. PLoS ONE 10(6): e0128444.
P. Krcah, Adaptation of Virtual Creatures to Different Environments Through Morphological Plasticity, in E. Tuci et al. (Eds.): SAB 2016, LNAI 9825, pp. 113–125, 2016
Rieffel, J., Knox, D., Smith, S. and Trimmer, B (2014) Growing and Evolving Soft Robots, Artificial Life 20 (1)
Rieffel, J., Valero-Cuevas, F. and Lipson, H. (2010) Morphological Communication: Exploiting Coupled Dynamics in a Complex Mechanical Structure to Achieve Locomotion. J. R. Soc. Interface
S. Doncieux, N. Bredeche, J.-B. Mouret and A.E. Eiben, Evolutionary robotics: what, why, and where to, Frontiers in Robotics and AI, 2(4) 2015
A.E. Eiben and J. Smith, From evolutionary computation to the evolution of things, Nature, 521:476-482,
J. E. Auerbach and J. C. Bongard, Environmental Influence on the Evolution of Morphological Complexity in Machines, PLoS Computational Biology, vol. 10, num. 1, p. e1003399, 2014.
Christensen, D.J.;Schultz, U.P.; Stoy, K., Fault-tolerant gait learning and morphology optimization of a polymorphic walking robot, Evolving Systems, 5(1):21-32, 2014.
A. P. Moczek, S. Sultan et al. (2011) The role of developmental plasticity in evolutionary innovation, Proc. R. Soc. B 278, 2705–2713
D.W. Pfennig, M. A. Wund, E. C. Snell-Rood, T. Cruickshank, C. D. Schlichting and A. P. Moczek, Phenotypic plasticity’s impacts on diversification and speciation, Trends in Ecology and Evolution 25 (2010) 459–467.