Simulation And Evaluation Of Multi-Agent System

Multi-agent system is a set of agents,include crowd,traffic flow,insect swarm,bird flock and so on.In a multi-agent system,collective behaviors emerge because of the interactions among agents.The applications of the simulation of multi-agent systems include VR therapy of phobk,traffic simulation in autonomous driving,urban design and planning,driving sim-ulators for education and digital entertainment,etc.Recently,researchers try to simulate the behaviors of multi-agent systems,and evaluate the results.However,some problems remain unsolved:1.How the group behavior is produced?What are the key factors?2.How to simulate when heterogeneous agents are mixed together;3.How to evaluate the similarity of motion of different trajectories of the same species.To address above issues,we have achieved the following results:1.This dissertation proposes a biologically inspired method for simulation of large-scale bird flocks.It uses information transfer network to describe the logical relationships between individuals to convey information,and a moving hash table to describe the spatial relationships between individuals.In practice,this approach can generate di:fferent bird behaviors,including dancing and large-scale aggregation.In addition,this model can capture plausible characteristics of the movement of the birds.2.This dissertation presents an information transfer network in which interactions last for a certain period of time.It then combines this network with a dynamic model of self-propelled particles.This dissertation observes that the resulting model that uses a stable information transfer network creates a bird flock with a more robust performance.It further shows that the time of information transfer in a flock grows logarithmically with its size and is proportional to the average response time of the birds.Moreover,it finds that the ranking curves displaying the order in wlich birds first perceive an external stimulus have similar shapes across different flocks.The results demonstrate that,beyond the traditional local,temporal interactions,the stable information transfer network serves as an efficient mechanism to model the emergence of large-scale collective behavior.3.This dissertation introduces a novel approach that combines physics-based simula-tion methods with data-driven techniques using an optimization-based formulation.This approach is general and can simulate heterogeneous agents corresponding to human crowds,traffic,vehicles,or combinations of different agents with varying dynamics.This method estimates motion states from real-world datasets that include information about position,velocity,and control direction.The optimization algorithm in this approach considers sev-eral constraints,including velocity continuity,collision avoidance,attraction,and direction control.To accelerate the computations,this method reduces the search space for both collision avoidance and optimal solution computation.This approach can simulate tens or hundreds of agents at interactive rates and it compares its accuracy with real-world datasets and prior algorithms.This method also performs user studies that evaluate the plausible behaviors generated by our algorithm and a user study that evaluates the plausibility of our algorithm via VR.4.This dissertation presents a data-driven evaluation method for the motions of insect swarms.This approach is based on pre-recorded insect trajectories.After presenting a novel evaluation metric and a statistical validation approach that takes into account various characteristics insect motions,this method evaluates well-known noise functions.Finally,this method combines Curl noise function with a dynamics model to generate realistic swarm simulations and emergent behaviors of flying insects.The combined model can simulate large swarms with a high performance.