A new system architecture for crowd simulation

2010 IEEE International Symposium on Parallel &Amp; Distributed Processing, Workshops and PHD Forum (IPDPSW)

2010

The current challenge for crowd simulations is the design and development of a scalable system that is capable of simulating the individual behavior of millions of complex agents populating large scale virtual worlds with a good frame rate. In order to overcome this challenge, this thesis proposes different improvements for crowd simulations. Concretely, we propose a distributed software architecture that can take advantage of the existing distributed and multicore architectures. In turn, the use of these distributed architectures requires partitioning strategies and workload balancing techniques for distributed crowd simulations. Also, these architectures allow the use of GPUs not only for rendering images but also for computing purposes. Finally, the design and implementation of distributed visual clients is another research topic that can help to overcome this challenge. Description of the ProblemCrowd simulation has become an essential tool for many virtual environment applications, and the extensive use of high quality virtual crowds is crucial for many virtual environment applications in education, training, and entertainment [1]. Crowd simulation can be considered as a special case of Virtual Environments where the avatars are intelligent agents instead of user-driven entities. Each of these agentbased entities can have its own goals, knowledge and behavior [2].On the one hand, crowd simulations must focus on rendering visually plausible images of the environment, requiring a high computational cost. On the other hand, complex agents must have autonomous behaviors, greatly increasing the computational cost as well. The sum of these requirements results in a computational cost that exponentially increases with the numbers of agents in the system, requiring a scalable design that can support huge amounts of agents (of different orders of magnitude) by simply adding more hardware. Thus, some proposals tackle crowd simulations as a particle system with different levels of details (eg:impostors) in order to reduce the computational cost [3] due to the graphical quality. Also, several proposals have been made to provide efficient and autonomous behaviors to crowd simulations [4]. The current challenge for these applications is the design and development of a scalable system that is capable of simulating the individual behavior of millions of complex agents populating large scale virtual worlds with a good frame rate.

Section: The Proposed Approach and Methodologymentioning

confidence: 99%

Performance improvements of real-time crowd simulations

Vigueras

Orduña

2010 IEEE International Symposium on Parallel &Amp; Distributed Processing, Workshops and PHD Forum (IPDPSW)

2010

“…We have implemented this module as a Cellular Automata (CA). The number and length of the paths computed can be adjusted in order to reduce the memory used by the CA [6]. On other hand, a low-level navigation module determines how the paths computed by the CA should be followed.…”

Section: Application Examplementioning

confidence: 99%

A GPU-Based Multi-agent System for Real-Time Simulations

Vigueras

Orduña

Advances in Intelligent and Soft Computing

2010

The huge number of cores existing in current Graphics Processor Units (GPUs) provides these devices with computing capabilities that can be exploited by distributed applications. In particular, these capabilites have been used in crowd simulations for enhancing the crowd rendering, and even for simulating continuum crowds. However, GPUs have not been used for simulating large crowds of complex agents, since these simulations require distributed architectures that can support huge amounts of agents. In this paper, we propose a GPU-based multi-agent system for crowd simulation. Concretely, we propose the use of an on-board GPU to implement some of the tasks that a distributed server for crowd simulations should perform. The huge number of cores in the GPU is used to simultaneously validate movement requests from different agents, greatly reducing the server response time. Since this task represents the critical data path, the use of this hardware significantly increases the parallelism achieved with respect to the implementation of the same distributed server on a CPU. An application example shows that the system can support agents with complex navigational behaviors.

“…local minimum or deadlocks) can appear. Beyond handwritten rules systems, other discrete techniques, as celular automata or multi-layer grids have been also used in these domains (Lozano et al, 2008) for representing different kind of navigational maps, as they can precompute important information for the agents when they have to plan and to follow their paths.…”

Section: Introductionmentioning

confidence: 99%

A Reinforcement Learning Approach for Multiagent Navigation

Martinez-Gil

Barber

Proceedings of the 2nd International Conference on Agents and Artificial Intelligence

et al. 2010

Abstract:This paper presents a Q-Learning-based multiagent system oriented to provide navigation skills to simulation agents in virtual environments. We focus on learning local navigation behaviours from the interactions with other agents and the environment. We adopt an environment-independent state space representation to provide the required scalability of such kind of systems. In this way, we evaluate whether the learned action-value functions can be transferred to other agents to increase the size of the group without loosing behavioural quality. We explain the learning process defined and the results of the collective behaviours obtained in a well-known experiment in multiagent navigation: evacuation through a door.