Parallel Programming with CUDA for RoboCup Rescue Simulation

The RoboCup Rescue Agent Simulation League (RSL) is a challenging competition to test multi-agent planning methods for disaster mitigation. The simulator creates a disaster scenario with three main tasks for the agents; firefighting, removing blockades on the road, and taking civilians to the refugees. Currently, the simulator operates on a small map of the cities. The whole simulator is designed to run sub-simulators (such as fire simulator and traffic simulator) in a distributed fashion over TCP/IP.

The aim of this project is to parallelize algorithms of sub-simulators using CUDA and JCuda SDK. The following tasks are to be carried out to realize this project.

• The analysis of sub-simulators to identify the opportunities to parallelize the algorithms. (For example, fire propagation can be calculated independently for every cell/buildings on the map)

• The implementation of JCuda algorithms for sub-simulators.

• The evaluation of the developed algorithms. (The detailed experimental analysis of the performance increase)

For further information

• http://roborescue.sourceforge.net/blog/

• https://developer.nvidia.com/cuda-downloads

• http://jcuda.org/

• You can contact Okan Aşık (asik.okangmail.com ())

Notes

• You will need a laptop/desktop computer having CUDA-enabled GPU.

• The project will be implemented mainly using Java programming language, but you might also implement the C/C++ CUDA kernel.