GPGPU Cost Function Generation for 3D Occupancy Volumes

A simulation of a GPU based cost-to-go function generator for 3D occupancy volumes. The volume is mapped to 2D so that it can be easily evaluated using a GLSL implementation of the GPGPU method of reduction using the ping-pong method. Just like the 2D occupancy grid implementation can evaluate a 1D wavefront of cells concurrently, this 3D version can evaluate a 2D surface of wavefront cells concurrently. (Please note that this is only simulation. The real implementation runs in the order of less than 20ms per complete cost-to-go function evaluation).

Category:

Science & Technology

Tags:

• GPGPU
• Dynamic Programming
• path planning
• cost-to-go function
• robotics