FAME-QP: A GPU-Accelerated Quadratic-Programming Solver for Real-Time Model Predictive Control
The authors report lower latency than competing GPU solvers while preserving solution accuracy and control performance.
FAME-QP solves dense quadratic programs arising in linear model predictive control. It combines constraint-dependent operator representations with fused GPU iterations replayed through CUDA graphs. Device calibration and problem-specific timing select the execution configuration, while cached operators support penalty selection and are updated lazily when problem matrices change. The reported results come from numerical benchmarks and simulated humanoid and quadrotor control experiments.
HF Daily Papers' note
FAME-QP solves dense quadratic programs arising in linear model predictive control. It combines constraint-dependent operator representations with fused GPU iterations replayed through CUDA graphs. Device calibration and problem-specific timing select the execution configuration, while cached operators support penalty selection and are updated lazily when problem matrices change. The reported results come from numerical benchmarks and simulated humanoid and quadrotor control experiments.
HF Daily Papers' note
score 4