Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.749496
Title: Agent-based pedestrian simulation on GPUs for use in Decision Support Systems
Author: Karmakharm, Twin
ISNI:       0000 0004 7233 873X
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Agent-based simulation of pedestrian crowds in public spaces can give insights into the potential congestion areas and space utilisation allowing new spaces to be better designed and existing ones to be more efficiently managed. This thesis investigates the components essential for a Graphics Processing Unit (GPU) based pedestrian simulation system that has the potential to be integrated into a real-time Decision Support System (DSS). Two navigation approaches are presented. A novel agent-based navigation grid approach uses a grid of agents to represent navigation information and for static obstacle avoidance. A case study of an urban environment shows it is a straightforward and efficient way of implementing navigation behaviour, allowing for the FLAME GPU agent-based framework to handle the necessary GPU optimisation. Then, a novel searchable and fully-resolved navigation graph approach is presented that allows pedestrian agents to make branching decisions and minor route changes in the case of congestion. Two cases of a shopping mall and a train station based on real environments show that navigation access times are good and memory use is two orders of magnitude less than the grid-based approach. A prototype pedestrian simulation software Concoursia is presented which inte-grates the searchable navigation graph approach and allows the authoring of environ-ments, visualisation, and collection metrics. Support for public transport services, queue agents and waiting behaviours are also implemented. The results show that the system can be used to create and simulate complex environments where pedestrians have a large number of navigation goals. Finally, a pedestrian multi-simulation system is presented that manages the sim-ulation on multiple machines equipped with GPUs. Multiple simulation instances are merged and run as a single simulation efficiently utilising the parallel architecture of the GPU. An initial trial has shown that the system is able to dispatch and run multiple simulations concurrently on multiple machines.
Supervisor: Maddock, Steve Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.749496  DOI: Not available
Share: