Accelerating Branch MPC with Two-Level Parallel Direct Solves on GPUs
The episode discusses a paper accelerating Branch Model Predictive Control (MPC) using two-level parallel direct solves on GPUs. The hosts explain how this method uses specific matrix structures to gain massive parallelism across scenarios and prediction horizons, directly speeding up the Cholesky factorization and triangular solve steps for the underlying linear system.