Nos partenaires



Accueil du site > English > Research Topics > Topic 1 - Information Analysis and Synthesis > VORTEX Team > VORTEX Team Research Activities

Artificial Life


Research Activities :

The VORTEX/ALIFE group tackles another aspect of computer graphics : the automatization of processes by the use of artificial life paradigms. This applies to shape generation, synthesis of Artificial Creatures in 3D Dynamic Environments and on Behavioral Animation. The use of Artificial Life paradigms is a commitment of this group since its first works that allows differenting from other teams while providing promising results. The group has achieved some major theoretical advances (Gene Regulatory Networks, Learning Classifiers Systems, Multi-Objective Optimization, etc) that led us to enlarge our scope. Therefore, these researches cover different applications fields : design of the morphology and the controller for virtual entities or characters, biology simulation of cells growing, artistic projects, design modular and/or reconfigurable robots, A.I. for serious games and multi-objective optimization.

Developed paradigms. Our approach deals with the breakthrough paradigm of “Morphogenetic Engineering” [14690]. We have developed during the last years a novel model based on Banzhaf’s original Genetic Regulatory Networks [12864]. This model allows generating by evolution the design of a GRN, which could then be used as controller [13878] or as starter in an artificial embryogeny simulation. In the field of agent control, this model appears competitive with other approaches such as NEAT or CPPN.

Behavioral animation. We have proposed new models for generating behaviors at several levels. For crowds, we propose an imitation based approach [11707] and a psychology-believable model based [13865]. At agent level we propose a new model for the automatic generation or behavior trees for NPS in serious games [15663] and the use of GRN as driver for AUV&cars (3rd place at GECCO 2013 car racing competition) or arbitrating between conflicting resources [13878].

Artificial creatures and morphogenetic engineering. In past work, we proposed an improved model of the Karl Sims approach [8001] ; we now have made a new step in the direction of automatic generation of virtual creatures from a single egg by proposing the use of GRN as controller for the growth of artificial simulated cells [11961]. This model has thus been applied in several fields : in robotics in the context of a collaboration with Brandeis U. Demo Lab [14513], in ressource layout optimization in collaboration with MIT’s ALFA group [14511]. The latter running collaboration leads to a model extension to tackle continuous environments.

Our main contribution concerns, a new artificial embryogeny model based on Artificial Regulatory Network published in the seminal book “Morphogenetic Engineering” [14690], a model and a simulator of the real Cell cycle in collaboration with the biologists of ITAV [14690], and for behavioral simulation, new A.I. approach (BPMN/MCTS) for complex health serious game.

Research project :

The ALIFE group main challenge is to automatically produce self-* complex artificial systems in 3D simulated environments. For that purpose it has to face intricate problems that blends up to date evolutionary algorithms, learning systems and multi level, multi purpose (3D w/o physics w/o chemistry, etc.) simulators. The blend between several simulators thus requires the group to reinforce its internal skills (theorical and HPC) and to gather high level collaborations. The last associate professor recruitement is a step in that direction. Our researches driven with strong theorical advances will evolve thru four main axes : behavioral animation, artificial creatures, morphogenetic engineering and new collaborations. We will pay a big attention to the interface between Mathematics, Biology and Computer Science that will provide a hotbed of creativity and collaborations opportunities.

Behavioral Simulation. The next main challenge is the control of learning and decision- making by a predictive algorithm such as the "Cortical Learning Algorithm”(HTM Jeff Hawkins). A first thesis will be presented this year, and we start collaboration with the SGRN to develop this axis.

Artificial Creatures. We have to deepen Artificial Regulatory Networks for Generative Developmental Systems and for Embryogeny and Organogeny. Thus, accurate physic simulators must be designed for the cells and for the simulated proteins in the environment. Our strategy to carry out that work, is to continue the collaboration with ITAV and to increase our participation to projects at the edge between biology and computer science. Moreover, novel collaborations with robotics labs such as DemoLab (Brandeis-U) or ISIR (UMR7222) will be established when the aim is to transfer our simulations into real robots.

Morphogenetic Engineering. This is a Breakthrough paradigm. The big challenge is to design a “complex systems” which develops according to its environment then performs a function. One big challenge is to control the
printing of such a “Creature” by a Cells 3D printer. That’s why we involved in the creation of the Fablab and we currently build together research projects for the next years.

New collaborations. The theorical advances achieved by the group leads to a diversification in the application field materialized by new collaborations. We are developing new projects in the field of simulation applied to biology (genomics, cell simulation, genotype/phenotype matching), ressources placing and analysis with both internal (APO team) and external partners (AMIS (UMR5288), MIT ALFA Group, ITAV (USR3505) IP3D team).

Research Contributions :

Behavioral Animation using GRN controllers
Artificial Genetic Regulatory Networks have initially been developed for artificial embryogeny applications. They are behind the cells differentiation mechanism underlaying several models. We show here how to use them as reactive controllers in a car driving simulator. This application has gained the 3rd prize for controllers at GECCO’s car driving comp in 2013.
On the left are shown a the sensors which are used as inputs for the grn controller. The output controls the steering, accelarating/braking. On the right is a picture of the VORTEX car in the simulator.
Genetic Programming and Evolvable Machines, Springer, Vol. 15 N. 2, june 2014.
Checkpoint Orientated Cell Cycle Simulation
This project aims at creating a biologically credible simulator for tumors growth in proliferation phase. It proposes a checkpoint oriented cell cycle simulator linked to a 3D simulator of both cell’s adhesion and nutriments diffusion.
3D simulated tumor growth. Colors represents the cell’s internal checkpoint. Black cells are dead.
Artificial Life, Lansing (USA), The MIT Press, p. 465-472, july 2012.
The SOMA Project (Self Organizing Multicellular Structures)
The SOMA project (Self Organizing MulticellulAr Structures) is a a new model for the development of artificial creatures from a single cell. The model aims to provide a more biologically plausible abstraction of the morphogenesis and the specialization process, which the organogenesis process follows. It is built upon three main elements : a cellular physics system that simulates division and intercellular adhesion dynamics, a simplified cell cycle offering to the cells the possibility to select actions such as division, quiescence, differentiation or apoptosis and, finally, a cell specialization mechanism quantifying the ability to perform different functions. An evolved artificial gene regulatory network is employed as a cell controller.
Screen capture of the organism toward the evolutionary process. After 20 generations, the organims organize nutritive cells (in orange) into a cluster and constantly renew a very few defensive cells (in blue) which position are not yet optimal. At generation 20, the organism produces multiple highly organized cluster with a central nutritive cell protected by a field of defensive ones. The energy repartition (showed on the right-hand side, the green the higher) is very ecient with strategy. Finally, the best strategy is to produce a bigger cluster that constantly produce new cells in order to escape from the center of the environment, where the particles are concentrated. The cells use morphogens to position themselves in the organism and drive it in a specific direction.
Artificial Life, New York, The MIT Press, july 2014.
IA for Serious Games
Since our first works in behavioral simulation we propose novel algorithms for AI agent control in virtual environments. This experience is now translated for the automatic generation of NPC behavior in Serious Games. In the 3DVOR Project, The Monte Carlo Tree Search is used as a search algorithm to explore a search space where every potential solution reflects a specific state of the game environment. Functions representing the interaction abilities of each character are provided to the algorithm to leap from one state to another. We show that the MCTS algorithm successfully manages to plan the actions for several virtual characters in a synchronized fashion, from the initial state to one or more desirable end states. Besides, we demonstrate the ability of this algorithm to fulfill a specific requirement of a learning game AI : guiding the non player characters to follow a predefined and constrained learning scenario and, if necessary, to adapt their decision to unexpected events in the simulation.
3DVOR decision system’s AI MCTS tree
IEEE Conference on Computational Intelligence and Games (GIG 2014), Dortmund, Germany, 2014.
WindFarm Layout Dynamic Optimization
This project proposes a novel approach to generate wind farm layouts. Their design is a complex process that involves many stakeholders and combines intensive computer simulation, optimization techniques and manual iterations. We propose the use of a a bio-inspired artificial-cell-based developmental model. The simulated environment is inoculated with a virtual cell. This cell can divide in order to populate the environment. When the development of the virtual organism is finished, all the cells are assumed to represent turbines to generate the final wind farm layout. The behavior of the cells is optimized by an evolutionary algorithm to generate an optimal layout. A fundamental advantage of this approach is that it scales up easily : once learned, the cell’s behavior can be reused in various environments without further learning.
WindFarm Growth
Genetic and Evolutionary Computation COnference (GECCO 2014), Vancouver, ACM, July 2014.

Back Top