GPU based Position Based Dynamics for Surgical Simulators.

Area of Science:

• Computer Graphics
• Scientific Simulation
• Medical Simulation

Background:

• Position-Based Dynamics (PBD) is a widely used method for simulating dynamic systems in computer graphics.
• Simulating volume rendering with linear deformation in virtual scenes remains a computational challenge.
• Interactive multi-modal surgical simulation requires efficient and fast physics calculations.

Purpose of the Study:

• To implement a Graphics Processing Unit (GPU)-based Position-Based Dynamics solver within the iMSTK open-source toolkit.
• To accelerate physics calculations for interactive surgical simulations.
• To evaluate the performance of the GPU-based PBD implementation against existing solvers.

Main Methods:

• Implemented GPU-based Position-Based Dynamics using NVIDIA's CUDA toolkit.
• Executed vector calculations on GPU kernels, ensuring thread safety to prevent data overwrites.
• Tested the implementation on two computers with varying specifications using consumer-grade GPUs.
• Maintained consistent mesh complexity (959 vertices, 2591 tetrahedral elements) across all tests.

Main Results:

• Achieved a speedup of nearly 10x for physics calculations compared to CPU-based methods.
• Reduced calculation time for a 128x128 simulation from 7900ms (CPU) to 820ms (GPU).
• Demonstrated the effectiveness of GPU acceleration for PBD in the context of surgical simulation.

Conclusions:

• The GPU-based Position-Based Dynamics implementation significantly accelerates physics computations for surgical simulations.
• This advancement enables more responsive and realistic virtual surgical training environments.
• The integration into iMSTK provides a valuable tool for rapid prototyping of interactive surgical simulators.