Technical Approach

1
GPU-Accelerated Route Planning for Computer
Generated Forces UNC-CH/PEOSTRI/RDECOM/SAIC
Start point
Terrain Features And Route Segments
Render Features (Once)
Feature Buffer (static)
Segments
Generated Route
Full Feature set
Features
Lake
Cull segment set against feature set (GPU)
Reduced segments
Features lakes, rivers, trees, buildings, etc.
Cull feature set against segment set (GPU)
River
End point
Reduced features and segments

• Issues Concerning Route Planning for Computer
  Generated Forces (CGF)
• High computational complexity
• Widely applicable in simulation development
• Can consume 50 of simulation time
• Intersection calculations bottleneck route
  planning

Cull feature set against Single segment (GPU)
Exact feature/segment Tests (CPU)
Results

• Current Status of Graphics Processing Units
  (GPUs)
• Integral part of modern computers
• Performance increases faster than Moores Law
• Additional computational power to assist CPU

• Optimizations
• Culling is performed using the GPUs occlusion
  query capability.
• Each successive step reduces the number of
  segments and features that are tested in the
  subsequent steps
• Final group tested by the CPU is minimized
• Conservatively expand feature and segment sizes
  to avoid inaccuracy in calculations

• Overview of GPU-Accelerated Route Planning
• Ability to test multiple route segments in
  parallel
• Queries all potential segments faster than
  individually testing on the CPU
• More efficient search of the large planning
  space
• System integrated with OneSAF
• Demonstrated 30-50x speedup in feature analysis
  computation
• Demonstrated 10x speed up of route planning in
  OneSAF on a single CPU-GPU machine

• Three Phase GPU-Based Culling Procedure
• The number of segments is reduced by culling
  them against the full feature set
• The number of features is reduced by culling
  them against the reduced set of segments
• The reduced feature set is culled against each
  individual segment in the reduced segment set

Participants
Dinesh Manocha (UNC-CH) Troy Dere (RDECOM)
Ming C. Lin (UNC-CH) Angel Rodriguez (RDECOM)
Russel Gayle (UNC-CH) Marlo Verdesca (SAIC)
David Knott (UNC-CH) LTC John Surdu (PEOSTRI) Maria Bauer (RDECOM) Jaeson Munro (SAIC) Eric Root (SAIC)