CA-AC-MPC: CUDA-Accelerated Actor-Critic Model Predictive Control

2026-05-27

AI Summary

This work addresses the high-latency bottleneck in Actor-Critic model predictive control (MPC) caused by repeatedly solving optimization problems during both training and inference. To overcome this challenge, the paper introduces, for the first time, a CUDA-accelerated differentiable MPC layer that efficiently parallelizes the forward and backward passes of the optimization procedure. By deeply integrating reinforcement learning with MPC, the proposed approach maintains near-optimal dynamic control performance while substantially reducing end-to-end computational latency. Evaluated on agile drone racing tasks, the system achieves state-of-the-art lap times and significantly shortens both training and inference durations, demonstrating the efficiency and practicality of the proposed architecture.

Abstract

In the literature, actor-critic model predictive control (AC-MPC) integrates MPC with reinforcement learning to enable high-performance control of complex dynamical systems. However, its differentiable MPC layer requires repeatedly solving an optimization problem in both the forward and backward passes, leading to substantial training and inference latency. This paper tackles this bottleneck introducing a CUDA-accelerated variant that significantly reduces end-to-end execution time while preserving the control performance of the baseline formulation. Simulation results on an agile drone racing task show that our approach achieves state-of-the-art lap times and near-limit dynamic behaviour with markedly reduced training and inference time.

Problem

• Actor-Critic Model Predictive Control
• Differentiable MPC
• Optimization Latency
• Training Time
• Inference Time

Innovation

• CUDA acceleration
• Actor-Critic
• Model Predictive Control
• Differentiable Optimization
• Reinforcement Learning