pith. sign in

arxiv: 1907.05464 · v1 · pith:KVEB2HOPnew · submitted 2019-07-11 · 📡 eess.SY · cs.SY

Integrated Offline and Online Optimization-Based Control in a Base-Parallel Architecture

Anahita Jamshidnejad , Gabriel Gomes , Alexandre M. Bayen , Bart De Schutter This is my paper

Pith reviewed 2026-05-24 22:44 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords model predictive controlreal-time controlnonlinear systemstraffic flow controlparallel optimizationwarm-start initializationsampling-time constraints
0
0 comments X

The pith

Base controllers tuned offline initialize parallel online optimizers that solve constrained nonlinear MPC problems inside fixed sampling times.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a control architecture that runs offline-tuned base controllers alongside several online optimization-based controllers operating in parallel. Each base controller supplies an initial guess for the online solvers, which must finish within the control sampling interval. In a highway traffic simulation the combined system produces the lowest total cost while keeping every computation inside the time limit and yielding per-vehicle costs nearest to those of full online MPC. A reader would care because the approach targets the practical barrier that prevents high-performance model-predictive control from running on fast nonlinear plants with many constraints.

Core claim

The architecture pairs base controllers, which are designed or optimized offline, with parallel online controllers that solve the full optimization problem; the base solutions serve as warm starts so that at least one online solver returns a usable input before the sampling deadline expires.

What carries the argument

The base-parallel architecture, where offline base controllers generate starting points for multiple parallel online optimizers constrained to a sampling-time budget.

If this is right

  • The architecture can be altered online by adding or removing controllers without redesigning the rest of the system.
  • In the traffic example the overall realized cost is lower than that of the other real-time methods tested.
  • The average cost per vehicle lies closer to the cost achieved by unlimited-time online MPC than the costs of the other online approaches considered.
  • All computations remain inside the given time budget for the nonlinear traffic model.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same initialization strategy could be tested on other fast nonlinear plants such as vehicle platoons or power converters to check whether the time-budget guarantee holds.
  • One could measure how solution quality scales when the number of parallel solvers is increased while the total wall-clock time is kept fixed.
  • The online flexibility to insert or delete controllers suggests an adaptive variant that activates extra solvers only when the current state indicates high constraint activity.

Load-bearing premise

The base controllers will consistently supply starting points that allow at least one parallel online optimizer to reach an acceptable solution inside the strict sampling-time budget for the target nonlinear system.

What would settle it

A run of the traffic simulation in which every parallel optimizer initialized from the base controllers returns an infeasible or markedly worse solution before the sampling deadline.

Figures

Figures reproduced from arXiv: 1907.05464 by Alexandre M. Bayen, Anahita Jamshidnejad, Bart De Schutter, Gabriel Gomes.

Figure 1
Figure 1. Figure 1: Offline tuned controller: parameters are tuned initially and may be updated regularly (τ is a counter for tuning sampling time steps, which occur less frequently than control sampling time steps counted by k). PSfrag replacements θ old(τ) θ up(τ) x m(k) u(k) offline tuned controller system tuning module training dataset representing desired behaviors realized in the past and desired future behaviors predic… view at source ↗
Figure 2
Figure 2. Figure 2: Optimal controller: open-loop, i.e., classical optimal controller (top plot) and closed-loop, i.e., MPC-based controller (bottom plot), with k0 the initial control sampling time step and k the progressive control sampling time step. Bold and regular letters are used for, respectively, vectors and scalars. For functions, a hat symbol is used on a small regular letter. The superscripts “m” and “p” indicate “… view at source ↗
Figure 3
Figure 3. Figure 3: Conventional MPC, parameterized MPC, and move block [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Applicability of various MPC-based approaches for r [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Base-parallel integrated control architecture for [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detailed illustration of the base-parallel integra [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Time-varying base controller vs. online optimizati [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The control input u(k) produced by the base controller should first be used to estimate the control input for the future sampling control time steps in the prediction time window. This estimation is done by a mathematical model of the controlled system and the uncontrolled external inputs. PSfrag replacements cell i−1 cell i cell i+1 oi−2(k s ) oi−1(k s ) oi(k s ) oi+1(k s ) on-ramp i off-ramp i di(k s ) s… view at source ↗
Figure 9
Figure 9. Figure 9: The main components and variables used in the ACTM. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Stretch of a single-lane highway used for the case st [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Demand profiles of the mainstream road (top plot) and [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The base-parallel integrated control architectur [PITH_FULL_IMAGE:figures/full_fig_p010_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The ANN-based mapping used as an implicit base contr [PITH_FULL_IMAGE:figures/full_fig_p010_13.png] view at source ↗
read the original abstract

We propose an integrated control architecture to address the gap that currently exists for efficient real-time implementation of MPC-based control approaches for highly nonlinear systems with fast dynamics and a large number of control constraints. The proposed architecture contains two types of controllers: base controllers that are tuned or optimized offline, and parallel controllers that solve an optimization-based control problem online. The control inputs computed by the base controllers provide starting points for the optimization problem of the parallel controllers, which operate in parallel within a limited time budget that does not exceed the control sampling time. The resulting control system is very flexible and its architecture can easily be modified or changed online, e.g., by adding or eliminating controllers, for online improvement of the performance of the controlled system. In a case study, the proposed control architecture is implemented for highway traffic, which is characterized by nonlinear, fast dynamics with multiple control constraints, to minimize the overall travel time of the vehicles, while increasing their total traveled distance within the fixed simulation time window. The results of the simulation show the excellent real-time (i.e., within the given time budget) performance of the proposed control architecture, with the least realized value of the overall cost function. Moreover, among the online control approaches considered for the case study, the average cost per vehicle for the base-parallel control approach is the closest to the online MPC-based controllers, which have excellent performance but may involve computation times that exceed the given time budget.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes a base-parallel control architecture combining offline-tuned or optimized base controllers with parallel online optimizers that use base outputs as starting points and run within a strict sampling-time budget. The architecture is demonstrated on a highway traffic MPC case study minimizing overall travel time (with secondary objective of maximizing total distance traveled), where simulation results are reported to show real-time compliance and the lowest realized overall cost among compared methods, with average per-vehicle cost closest to full online MPC.

Significance. If the parallel optimizers consistently return feasible near-optimal solutions from base-controller initializations inside every sampling interval, the approach offers a flexible, modifiable way to obtain near-MPC performance for nonlinear fast systems under tight real-time constraints while permitting online addition or removal of controllers. The traffic application is a relevant domain for such methods.

major comments (2)
  1. [Case study results] Case study (simulation results): aggregate cost and timing numbers are presented without per-step solver status (feasible/infeasible/timeout), number of parallel instances executed, or confirmation that at least one optimizer always produced an acceptable iterate before the budget expired. These data are load-bearing for the claim of 'excellent real-time performance' and 'least realized value of the overall cost function'.
  2. [Proposed architecture] Architecture description: no convergence-rate bounds, feasibility-recovery procedure, or analysis of how base-controller starts affect the probability of reaching acceptable solutions within the time limit are supplied for the nonlinear traffic dynamics. This directly underpins the weakest assumption that parallel instances will reliably improve upon the base output.
minor comments (1)
  1. [Abstract] The abstract states that the base-parallel approach yields 'the least realized value of the overall cost function' and is 'the closest to the online MPC-based controllers' without quoting the numerical values or identifying the exact comparator set.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the presentation of the results and architecture.

read point-by-point responses

  1. Referee: [Case study results] Case study (simulation results): aggregate cost and timing numbers are presented without per-step solver status (feasible/infeasible/timeout), number of parallel instances executed, or confirmation that at least one optimizer always produced an acceptable iterate before the budget expired. These data are load-bearing for the claim of 'excellent real-time performance' and 'least realized value of the overall cost function'.

    Authors: We agree that per-step solver diagnostics would provide stronger support for the real-time claims. In the revised manuscript we will add a new table (or subsection) in the case study that reports, for every sampling instant: the number of parallel instances launched, the solver exit status (feasible/infeasible/timeout) for each instance, and explicit confirmation that at least one acceptable iterate was returned before the time budget expired. These data will be generated from the same simulation runs already performed. revision: yes

  2. Referee: [Proposed architecture] Architecture description: no convergence-rate bounds, feasibility-recovery procedure, or analysis of how base-controller starts affect the probability of reaching acceptable solutions within the time limit are supplied for the nonlinear traffic dynamics. This directly underpins the weakest assumption that parallel instances will reliably improve upon the base output.

    Authors: The architecture is presented as a practical, modular framework rather than a theoretical guarantee. We will therefore add a short discussion subsection that (i) describes a simple feasibility-recovery rule (fall back to the best base-controller output if no parallel optimizer returns a feasible point), (ii) reports empirical statistics from the traffic simulations on how often the base initialization led to improved feasible solutions within the allotted time, and (iii) clarifies that formal convergence-rate bounds are outside the scope of the work because the parallel solvers are black-box nonlinear programs. These additions will make the practical reliability of the assumption explicit without claiming theoretical guarantees. revision: partial

Circularity Check

0 steps flagged

No circularity: architecture proposal evaluated via external simulation benchmarks

full rationale

The paper proposes a base-parallel control architecture (offline-tuned base controllers supplying warm starts to parallel online optimizers) and evaluates it through case-study simulations on a highway traffic model. No derivation chain, equation, or performance claim reduces by construction to a fitted parameter, self-citation, or ansatz defined inside the paper itself. The reported real-time compliance and cost values are obtained from an external simulator and are therefore falsifiable independently of the architecture description.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities beyond the high-level description of base and parallel controllers.

pith-pipeline@v0.9.0 · 5800 in / 1123 out tokens · 17134 ms · 2026-05-24T22:44:43.400753+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    Maciejowski, Predictive Control with Constraints

    J. Maciejowski, Predictive Control with Constraints . London, UK: Prentice Hall, 2002

    work page 2002

  2. [2]

    Towards online model predic tive control on a programmable logic controller: Practical cons iderations,

    B. Huyck, H. J. Ferreau, M. Diehl, J. De Brabanter, J. F. M. V an Impe, B. De Moor, and F. Logist, “Towards online model predic tive control on a programmable logic controller: Practical cons iderations,” Mathematical Problems in Engineering , vol. 2012, pp. 1–20, 2012

    work page 2012

  3. [3]

    O nline model predictive control of industrial processes using low level control hard- ware: A pilot-scale distillation column case study,

    B. Huyck, J. De Brabanter, J. F. V an Impe, and F. Logist, “O nline model predictive control of industrial processes using low level control hard- ware: A pilot-scale distillation column case study,” Control Engineering Practice, vol. 28, pp. 34–48, 2014

    work page 2014

  4. [4]

    Fast model predictive control using online opti- mization,

    Y . Wang and S. Boyd, “Fast model predictive control using online opti- mization,” IEEE Transactions on Control Systems Technology , vol. 18, no. 2, pp. 267–278, 2010. 13

    work page 2010

  5. [5]

    An auto-generate d real- time iteration algorithm for nonlinear MPC in the microseco nd range,

    B. Houska, H. J. Ferreau, and M. Diehl, “An auto-generate d real- time iteration algorithm for nonlinear MPC in the microseco nd range,” Automatica, vol. 47, no. 10, pp. 2279–2285, 2011

    work page 2011

  6. [6]

    The explicit linear quadratic regulator for constrained systems,

    A. Bemporad, M. Morari, V . Dua, and E. N. Pistikopoulos, “ The explicit linear quadratic regulator for constrained systems,” Automatica, vol. 38, no. 1, pp. 3–20, 2002

    work page 2002

  7. [7]

    On multi-parametric nonlinear program ming and explicit nonlinear model predictive control,

    T. A. Johansen, “On multi-parametric nonlinear program ming and explicit nonlinear model predictive control,” in 41st IEEE Conference on Decision and Control , Las V egas, USA, December 2002, pp. 2768– 2773

    work page 2002

  8. [8]

    An algorithm for approximat e mul- tiparametric convex programming,

    A. Bemporad and C. Filippi, “An algorithm for approximat e mul- tiparametric convex programming,” Computational Optimization and Applications, vol. 35, no. 1, pp. 87–108, 2006

    work page 2006

  9. [9]

    Alessio and A

    A. Alessio and A. Bemporad, Nonlinear Model Predictive Control Lecture Notes in Control and Information Sciences . Berlin, Germany: Springer, 2009, vol. 384, ch. A Survey on Explicit Model Pred ictive Control

    work page 2009

  10. [10]

    Real-time s uboptimal model predictive control using a combination of explicit MP C and online optimization,

    M. N. Zeilinger, C. N. Jones, and M. Morari, “Real-time s uboptimal model predictive control using a combination of explicit MP C and online optimization,” IEEE Transactions on Automatic Control , vol. 56, no. 7, pp. 1524–1534, 2011

    work page 2011

  11. [11]

    A predictive traffic controller for sustainable m obility us- ing parameterized control policies,

    S. K. Zegeye, B. De Schutter, J. Hellendoorn, E. A. Breun esse, and A. Hegyi, “A predictive traffic controller for sustainable m obility us- ing parameterized control policies,” IEEE Transactions on Intelligent Transportation Systems, vol. 13, no. 3, pp. 1420–1429, 2012

    work page 2012

  12. [12]

    Parameterized infinite -horizon model predictive control for linear time-invariant systems with input and state constraints,

    M. Muehlebach and R. D’Andrea, “Parameterized infinite -horizon model predictive control for linear time-invariant systems with input and state constraints,” in American Control Conference (ACC), Boston, USA, July 2016, pp. 2669–2674

    work page 2016

  13. [13]

    Move blocking strategies in receding horizon control,

    R. Cagienard, P . Grieder, E. C. Kerrigan, and M. Morari, “Move blocking strategies in receding horizon control,” Journal of Process Control , vol. 17, no. 6, pp. 563–570, 2007

    work page 2007

  14. [14]

    Optimal move blocking stra tegies for model predictive control,

    R. C. Shekhar and C. Manzie, “Optimal move blocking stra tegies for model predictive control,” Automatica, vol. 61, pp. 27–34, 2015

    work page 2015

  15. [15]

    Gr¨ une and J

    L. Gr¨ une and J. Pannek, Nonlinear Model Predictive Control: Theory and Algorithms . London, UK: Springer-V erlag, 2011

    work page 2011

  16. [16]

    Suboptimal explicit reced ing horizon con- trol via approximate multiparametric quadratic programmi ng,

    A. Bemporad and C. Filippi, “Suboptimal explicit reced ing horizon con- trol via approximate multiparametric quadratic programmi ng,” Journal of Optimization Theory and Applications , vol. 117, no. 1, pp. 9–38, 2003

    work page 2003

  17. [17]

    Model predi ctive control of nonlinear systems: Computational burden and stability,

    W. H. Chen, D. J. Ballance, and J. O’Reilly, “Model predi ctive control of nonlinear systems: Computational burden and stability, ” IEE Proceed- ings - Control Theory and Applications , vol. 147, no. 4, pp. 387–394, 2000

    work page 2000

  18. [18]

    Fast NMPC: A reality-steered paradigm: Key properties of fast NMPC algorithms,

    M. Alamir, “Fast NMPC: A reality-steered paradigm: Key properties of fast NMPC algorithms,” in European Control Conference , Strasbourg, France, June 2014, pp. 2472–2477

    work page 2014

  19. [19]

    From linear to nonlinear MPC: Bridging the gap via the real-time i teration,

    S. Gros, M. Zanon, R. Quirynen, A. Bemporad, and M. Diehl , “From linear to nonlinear MPC: Bridging the gap via the real-time i teration,” International Journal of Control , pp. 1–19, 2016

    work page 2016

  20. [20]

    Optimal freeway ramp meterin g using the asymmetric cell transmission model,

    G. Gomes and R. Horowitz, “Optimal freeway ramp meterin g using the asymmetric cell transmission model,” Transportation Research Part C: Emerging Technologies, vol. 14, no. 4, pp. 244–262, 2006

    work page 2006

  21. [21]

    Met hodolog- ical calibration of the cell transmission model,

    L. Mu˜ noz, X. Sun, D. Sun, G. Gomes, and R. Horowitz, “Met hodolog- ical calibration of the cell transmission model,” in American Control Conference, Boston, USA, July 2004, pp. 798–803

    work page 2004

  22. [22]

    A LINEA: A local feedback-based control law for on-ramp metering,

    M. Papageorgiou, H. Hadj-Salem, and J. Blosseville, “A LINEA: A local feedback-based control law for on-ramp metering,” Transportation Research Rescord, vol. 1, no. 1320, pp. 58–67, 1991

    work page 1991

  23. [23]

    Modeling and real-time control of tr affic flow on the southern part of boulevard Peripherique in Paris: Part II: C oordinated on-ramp metering,

    M. Papageorgiou, “Modeling and real-time control of tr affic flow on the southern part of boulevard Peripherique in Paris: Part II: C oordinated on-ramp metering,” Transportation Research Part A , vol. 24A, no. 5, pp. 361–370, 1990

    work page 1990

  24. [24]

    Model pre dictive control for optimal coordination of ramp metering and variable spee d limits,

    A. Hegyi, B. De Schutter, and H. Hellendoorn, “Model pre dictive control for optimal coordination of ramp metering and variable spee d limits,” Transportation Research Part C, vol. 13, no. 3, pp. 185–209, Jun. 2005

    work page 2005

  25. [25]

    Efficient freeway MPC by parameterization of ALINEA and a sp eed- limited area,

    G. van de Weg, A. Hegyi, S. Hoogendoorn, and B. De Schutte r, “Efficient freeway MPC by parameterization of ALINEA and a sp eed- limited area,” IEEE Transactions on Intelligent Transportation Systems , vol. 20, no. 1, pp. 16–29, 2019. Anahita Jamshidnejad received the PhD degree in Systems and Control from the Delft University of Technology, the Netherland...

    work page 2019

  26. [26]

    He is Senior Editor of the IEEE Transactions on Intelligent Transportation Systems

    is a full professor and head of department at the Delft Center for Systems and Control of Delft University of Technology in Delft, The Netherlands. He is Senior Editor of the IEEE Transactions on Intelligent Transportation Systems. His current re- search interests include intelligent transportation and infrastructure systems, hybrid systems, and multi- le...

    work page