MPCGPU: Real-Time Nonlinear Model Predictive Control through Preconditioned Conjugate Gradient on the GPU

TL;DR

MPCGPU is a GPU-based real-time NMPC solver that significantly accelerates trajectory optimization, enabling kilohertz control rates for robotic manipulators by leveraging a custom preconditioned conjugate gradient method.

Contribution

This work introduces MPCGPU, a novel GPU-accelerated NMPC solver using an optimized PCG method, achieving faster computation and scalability over CPU-based solvers.

Findings

MPCGPU achieves at least 10x speedup over CPU solvers.

Enables control rates up to kilohertz for robotic manipulators.

Scales to longer trajectories with improved real-time performance.

Abstract

Nonlinear Model Predictive Control (NMPC) is a state-of-the-art approach for locomotion and manipulation which leverages trajectory optimization at each control step. While the performance of this approach is computationally bounded, implementations of direct trajectory optimization that use iterative methods to solve the underlying moderately-large and sparse linear systems, are a natural fit for parallel hardware acceleration. In this work, we introduce MPCGPU, a GPU-accelerated, real-time NMPC solver that leverages an accelerated preconditioned conjugate gradient (PCG) linear system solver at its core. We show that MPCGPU increases the scalability and real-time performance of NMPC, solving larger problems, at faster rates. In particular, for tracking tasks using the Kuka IIWA manipulator, MPCGPU is able to scale to kilohertz control rates with trajectories as long as 512 knot points. This is driven by a custom PCG solver which outperforms state-of-the-art, CPU-based, linear system solvers by at least 10x for a majority of solves and 3.6x on average.