Agent-based Simulation

Janus: an environment for implementation and simulation of multi-agent and holonic systems

Our team proposes an environment for implantation and simulation of multiagent holonic systems. This platform named Janus is based upon an holonic organisational metamodel and provides base tools to management the execution and the deployment of multiagent systems (message transport, scheduling, network support...) and of holonic systems (holon composition). This platform is designed for real time performances to be used by the 3D multiagent simulation projects. And thanks to its metamodel Janus is able to cover the implementation stage of the ASPECS methodology. The Janus platform is jointly developed with the (es) Centro de Investigación de Tecnologías Avanzadas de Tucumán.

See Janus for implementation platform.

Multi-view and Multilevel Environment

Our team aims at providing a complete agent-based simulator for large-scale complex systems (urban systems, buildings...). Because of the intrinsic nature of these systems, the simulation models should be split in two parts: the environment model and the simulated entity models. The environment model named JaSIM is containing all the simulated “physical representations” of the intelligent entities and all the objects required for there behaviors. Its goal is to maintain a set of world rules according to the agent’s actions and to provide a set of perceptions to all the agents (based on the Influence/Reaction model). We propose a multi-view environment model composed of several classical spatial dimension points of view (1D, 1.5D, 2D, 2.5D and 3D) and the corresponding mapping function to the agent’s spaces. The environment model and its implementation are designed to be very efficient (real-time as nearest). But to enlarge the size of the simulated environments we propose a multilevel model (for the 3D view for now) which permits to aggregate agent’s perceptions and actions according to spatial and efficiency constraints. This environment model was successfully applied on the simulation of a big European rock festival named Eurockéennes and on airport hall simulations. One of our industrial partner is using JaSIM environment as the basis of its application.

VIVUS platform includes a simplified version of environment model from JaSIM. It implements an 3D environment based on physics rules and implemented with PhysX engine.

See JaSIM and VIVUS for implementation platforms.

Multilevel Simulation Model

In the same context as the previous point, our team proposes an agent-based model based on multilevel simulation to be able to simulate a large amount of entities. This approach aims to group the agents into self-similar entities – or holons – when they have similar indicators. Our approach uses energy indicators to evaluate the accuracy of the simulation and when to merge agents into groups and split a group into agents. This model was successfully applied on pedestrian simulation.

See JaSIM for implementation platform.

Automatic Generation of 3D Environment Models

In the context of large-scale simulations realistic 3D models are used as entry data to generate the environment model and are displayed in a 3D viewer to help the simulation user to understand and manipulate the simulation parameters. But it is very difficult to manually build and design such 3D models is an efficient way. That why our team proposes a set of algorithms and tools to automatically generate the 3D models from different data sources: geographical information systems (GIS), architectural plans, photos... The tools extract the 3D meshes from the GIS and the architectural plans and refine them with other sources such as airplane photos. The generated models are assumed to have enough quality to be easily and efficiently update by a 3D designer.

See Live3D engine for implementation platform.

Activity Reports

Retrieved from ""
This page was last modified on 17 April 2013, at 08:26. This page has been accessed 41,751 times.
Copyright 2010-2020 © Laboratoire Connaissance et Intelligence Artificielle Distribuées - Université Bourgogne Franche-Comté - Privacy policy