Publications Neural Network Control
Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic
2026 American Control Conference (ACC 2026), May 2026, New Orleans, LA, USA
Abstract
We present a two-stage approach for learning stability-certified neural controllers that achieves a reduction of up to ∼95% in training time compared to the state-of-the-art baseline, which introduced monotonic neural Lyapunov architectures. Our method combines monotonic neural Lyapunov functions with fulfillment priority logic (FPL) to efficiently initialize controllers before formal verification. Traditional approaches for jointly learning controllers and neural Lyapunov functions require computationally expensive mixed-integer linear programming (MILP) or satisfiability modulo theory (SMT) solvers at each training iteration, often taking several hours to converge. We address this bottleneck by leveraging FPL to perform early joint initialization of the controller and Lyapunov networks. Building on the monotonic neural network architecture from the baseline, which guarantees non-negativity and a unique global minimum by construction, our method focuses on efficiently satisfying the remaining property of decreasing along trajectories. Existing works focus on maximizing the region of attraction/convergence of the learned controller. In contrast, leveraging FPL allows us to (1) increase learning efficiency substantially and (2) focus on complementary performance metrics, such as convergence rate and control effort minimization, thereby adding significant specification flexibility. In this paper, we encode an approximate Lyapunov-decrease condition in FPL to pre-train the controller and Lyapunov networks, then apply a MILP-based verification/refinement step. This decouples efficient learning from certificate enforcement and allows the FPL specification to include auxiliary objectives (e.g., convergence rate and control effort), whose influence persists through the final MILP pass. The resulting controllers converge rapidly while admitting formal Lyapunov certificates on standard nonlinear control benchmarks. 1
Code and hardware artifacts
Cite this paper
@inproceedings{FPL_Lyap_ACC26,
title = {{Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic}},
author = {Abdelgawad, Abdelrahman and El Mabsout, Bassel and Wang, Zili and Mancuso, Renato and Andersson, Sean B. and Tron, Roberto},
booktitle = {2026 American Control Conference (ACC 2026)},
year = 2026,
month = may,
address = {New Orleans, LA, USA},
url = {https://cs-people.bu.edu/rmancuso/files/papers/FPL_Lyap_ACC26.pdf}
}
TY - CONF AU - Abdelgawad, Abdelrahman AU - El Mabsout, Bassel AU - Wang, Zili AU - Mancuso, Renato AU - Andersson, Sean B. AU - Tron, Roberto TI - Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic T2 - 2026 American Control Conference (ACC 2026) PY - 2026 DA - 2026/05// CY - New Orleans, LA, USA AB - We present a two-stage approach for learning stability-certified neural controllers that achieves a reduction of up to ∼95% in training time compared to the state-of-the-art baseline, which introduced monotonic neural Lyapunov architectures. Our method combines monotonic neural Lyapunov functions with fulfillment priority logic (FPL) to efficiently initialize controllers before formal verification. Traditional approaches for jointly learning controllers and neural Lyapunov functions require computationally expensive mixed-integer linear programming (MILP) or satisfiability modulo theory (SMT) solvers at each training iteration, often taking several hours to converge. We address this bottleneck by leveraging FPL to perform early joint initialization of the controller and Lyapunov networks. Building on the monotonic neural network architecture from the baseline, which guarantees non-negativity and a unique global minimum by construction, our method focuses on efficiently satisfying the remaining property of decreasing along trajectories. Existing works focus on maximizing the region of attraction/convergence of the learned controller. In contrast, leveraging FPL allows us to (1) increase learning efficiency substantially and (2) focus on complementary performance metrics, such as convergence rate and control effort minimization, thereby adding significant specification flexibility. In this paper, we encode an approximate Lyapunov-decrease condition in FPL to pre-train the controller and Lyapunov networks, then apply a MILP-based verification/refinement step. This decouples efficient learning from certificate enforcement and allows the FPL specification to include auxiliary objectives (e.g., convergence rate and control effort), whose influence persists through the final MILP pass. The resulting controllers converge rapidly while admitting formal Lyapunov certificates on standard nonlinear control benchmarks. 1 ER -
%0 Conference Paper %A Abdelgawad, Abdelrahman %A El Mabsout, Bassel %A Wang, Zili %A Mancuso, Renato %A Andersson, Sean B. %A Tron, Roberto %T Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic %B 2026 American Control Conference (ACC 2026) %D 2026 %C New Orleans, LA, USA %X We present a two-stage approach for learning stability-certified neural controllers that achieves a reduction of up to ∼95% in training time compared to the state-of-the-art baseline, which introduced monotonic neural Lyapunov architectures. Our method combines monotonic neural Lyapunov functions with fulfillment priority logic (FPL) to efficiently initialize controllers before formal verification. Traditional approaches for jointly learning controllers and neural Lyapunov functions require computationally expensive mixed-integer linear programming (MILP) or satisfiability modulo theory (SMT) solvers at each training iteration, often taking several hours to converge. We address this bottleneck by leveraging FPL to perform early joint initialization of the controller and Lyapunov networks. Building on the monotonic neural network architecture from the baseline, which guarantees non-negativity and a unique global minimum by construction, our method focuses on efficiently satisfying the remaining property of decreasing along trajectories. Existing works focus on maximizing the region of attraction/convergence of the learned controller. In contrast, leveraging FPL allows us to (1) increase learning efficiency substantially and (2) focus on complementary performance metrics, such as convergence rate and control effort minimization, thereby adding significant specification flexibility. In this paper, we encode an approximate Lyapunov-decrease condition in FPL to pre-train the controller and Lyapunov networks, then apply a MILP-based verification/refinement step. This decouples efficient learning from certificate enforcement and allows the FPL specification to include auxiliary objectives (e.g., convergence rate and control effort), whose influence persists through the final MILP pass. The resulting controllers converge rapidly while admitting formal Lyapunov certificates on standard nonlinear control benchmarks. 1
[
{
"id": "FPL_Lyap_ACC26",
"type": "paper-conference",
"title": "Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic",
"author": [
{
"family": "Abdelgawad",
"given": "Abdelrahman"
},
{
"family": "El Mabsout",
"given": "Bassel"
},
{
"family": "Wang",
"given": "Zili"
},
{
"family": "Mancuso",
"given": "Renato"
},
{
"family": "Andersson",
"given": "Sean B."
},
{
"family": "Tron",
"given": "Roberto"
}
],
"container-title": "2026 American Control Conference (ACC 2026)",
"issued": {
"date-parts": [
[
2026,
5
]
]
},
"publisher-place": "New Orleans, LA, USA",
"abstract": "We present a two-stage approach for learning stability-certified neural controllers that achieves a reduction of up to ∼95% in training time compared to the state-of-the-art baseline, which introduced monotonic neural Lyapunov architectures. Our method combines monotonic neural Lyapunov functions with fulfillment priority logic (FPL) to efficiently initialize controllers before formal verification. Traditional approaches for jointly learning controllers and neural Lyapunov functions require computationally expensive mixed-integer linear programming (MILP) or satisfiability modulo theory (SMT) solvers at each training iteration, often taking several hours to converge. We address this bottleneck by leveraging FPL to perform early joint initialization of the controller and Lyapunov networks. Building on the monotonic neural network architecture from the baseline, which guarantees non-negativity and a unique global minimum by construction, our method focuses on efficiently satisfying the remaining property of decreasing along trajectories. Existing works focus on maximizing the region of attraction/convergence of the learned controller. In contrast, leveraging FPL allows us to (1) increase learning efficiency substantially and (2) focus on complementary performance metrics, such as convergence rate and control effort minimization, thereby adding significant specification flexibility. In this paper, we encode an approximate Lyapunov-decrease condition in FPL to pre-train the controller and Lyapunov networks, then apply a MILP-based verification/refinement step. This decouples efficient learning from certificate enforcement and allows the FPL specification to include auxiliary objectives (e.g., convergence rate and control effort), whose influence persists through the final MILP pass. The resulting controllers converge rapidly while admitting formal Lyapunov certificates on standard nonlinear control benchmarks. 1"
}
]
A. Abdelgawad, B. El Mabsout, Z. Wang, R. Mancuso, S. B. Andersson, and R. Tron, “Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic,” in 2026 American Control Conference (ACC 2026), May. 2026.
Abdelgawad, A., El Mabsout, B., Wang, Z., Mancuso, R., Andersson, S. B., & Tron, R. (2026). Accelerating Lyapunov-Stable Neural Control using Fulfillment Priority Logic. In 2026 American Control Conference (ACC 2026).