[Paper Review] acados: a modular open-source framework for fast embedded optimal control

Robin Verschueren, Gianluca Frison+8 more|arXiv (Cornell University)|Oct 30, 2019

Advanced Control Systems Optimization19 citations

TL;DR

acados is a modular, open-source framework for fast embedded optimal control, implementing efficient nonlinear model predictive control (NMPC) and moving horizon estimation (MHE) solvers using high-performance linear algebra via BLASFEO. It enables real-time performance on embedded CPUs with sub-10ms computation times, achieving faster and more robust NMPC than state-of-the-art methods through exact-Hessian SQP with convexification and structure-exploiting integrators.

ABSTRACT

This paper presents the acados software package, a collection of solvers for fast embedded optimization intended for fast embedded applications. Its interfaces to higher-level languages make it useful for quickly designing an optimization-based control algorithm by putting together different algorithmic components that can be readily connected and interchanged. Since the core of acados is written on top of a high-performance linear algebra library, we do not sacrifice computational performance. Thus, we aim to provide both flexibility and performance through modularity, without the need to rely on automatic code generation, which facilitates maintainability and extensibility. The main features of acados are: efficient optimal control algorithms targeting embedded devices implemented in C, linear algebra based on the high-performance BLASFEO library, user-friendly interfaces to Matlab and Python, and compatibility with the modeling language of CasADi. acados is free and open-source software released under the permissive BSD 2-Clause license.

Motivation & Objective

• To address the performance-flexibility trade-off in embedded optimal control software by avoiding automatic code generation.
• To enable real-time NMPC and MHE on resource-constrained embedded systems with millisecond computation times.
• To provide a high-performance, extensible, and maintainable framework for embedded optimization without sacrificing computational efficiency.
• To support flexible algorithm composition through modular components and interfaces to Python and MATLAB.
• To demonstrate the feasibility and efficiency of exact-Hessian SQP methods in real-time applications via hardware-in-the-loop experiments.

Proposed method

• Uses BLASFEO as a high-performance linear algebra backend to accelerate core computations without code generation.
• Implements structure-exploiting integrators (GNSF-IRK) for efficient multiple shooting in NMPC and MHE.
• Introduces Hessian convexification to enable exact-Hessian SQP methods with minimal computational overhead.
• Supports partial condensing and variable state/control dimensions per shooting interval for modeling flexibility.
• Provides interfaces to CasADi for automatic differentiation and modeling, and to Python and MATLAB for rapid prototyping.
• Employs HPIPM as the QP solver, optimized for the problem structure and hardware targets.

Experimental results

Research questions

• RQ1 Can a modular, open-source framework achieve real-time performance in embedded optimal control without relying on automatic code generation?
• RQ2 How does the use of BLASFEO as a linear algebra backend compare to generated code in terms of speed and memory efficiency for embedded optimization?
• RQ3 Can exact-Hessian SQP methods be made practical for real-time NMPC through convexification and efficient linear algebra?
• RQ4 What is the achievable computation time for NMPC on industrial embedded platforms like dSPACE MicroAutoboxII?
• RQ5 How does acados perform in comparison to other embedded optimization packages in terms of speed and robustness?

Key findings

• acados achieves computation times under 10 ms on the dSPACE MicroAutoboxII (900 MHz PowerPC), which is 5× faster than the system’s 50 ms sampling time.
• The framework demonstrates real-time feasibility of exact-Hessian SQP methods via Hessian convexification, which adds only a small cost per iteration while enabling global convergence.
• Hardware-in-the-loop experiments show stable, oscillation-free tracking of boost pressure in engine control with active constraints and reference changes.
• Computation times spike at constraint activation or reference jumps, but remain within the real-time requirement despite dynamic changes.
• Performance on the dSPACE platform is about three times slower than on a desktop i7, yet still within the real-time window, demonstrating embeddability.
• acados outperforms state-of-the-art embedded solvers in speed and robustness, particularly in handling nonlinear dynamics and constraints with high accuracy.