கணினி பொறியியல் & தகவல் தொழில்நுட்ப இதழ்

SiMAMT: An Interactive 3D Graphical Simulation Environment for StrategyBased Multi-Agent Multi-Team Systems

Michael Franklin

Multi-agent multi-team environments are complicated and complex. The normal approach is to simplify the structure by using a single policy for each agent, such as in swarming or flocking algorithms. While this type of simulation environment may provide multiple agents working within the system, their interactions are singledimensional and their group behavior minimal. SiMAMT, in contrast, is a hierarchical, strategy-based approach that provides large-scale, complex strategic initiatives realized by independent intelligent single agents. These agents are independent because they have their own talents, skills, abilities, and behaviors that are influenced by the commands given to them from the layer above (e.g., the team). These agents can all have their own behaviors, or several could have similar behaviors, or entire teams could share one behavior, depending on the scenario. Further, SiMAMT utilizes strategy-based behaviors at every level, so the players are influenced by the team’s strategy, the teams are influenced by the unit’s strategy, the units are influenced by the battalions’ strategy, etc. Whichever hierarchical structure the environment needs — sports, military, organizational, etc. It can be supported by the SiMAMT system. The simulation environment provides the 3D visual environment to view the progress of the simulation from both an overall perspective and from a first-person perspective from each agent. This combination view provides insight into how each layer of the structural hierarchy is performing agents, teams, overall interaction, etc. Additionally, it provides overall views of the strategy that each team is using, each agent’s behavior, and the overlaps of both. The simulation also provides statistics as the simulation is running to relay observations, transitions, most likely strategies in play (the SiMAMT framework provides strategic inference to determine the most likely strategy being employed by the other teams in the environment), and overall simulation results. Overall, the goal of the simulation is to allow multi-agent teams to perform strategically in interactive time while performing strategy-inference to improve their performance. The SiMAMT simulation achieves this goal, and this will be demonstrated in the experiments.