2025/02/13 by Tingli Hu, Hu, Tingli, Sami Haddadin +1
Computer Science · Engineering · #Adaptive Control of Nonlinear Systems #Adaptive Dynamic Programming Control #Optimization and Variational Analysis #math.OC
paper · pdf · doi:10.48550/arxiv.2502.09322
openalex publication_date 2025/02/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/30
The real-time barrier in optimal control of nonlinear dynamical systems remains a longstanding limitation across science, engineering, and economics. Existing approaches rely on iterative optimization and therefore cannot compute optimal control actions directly within physical time for complex, high-dimensional systems. Here we introduce the Dual Cost-Constraint projection (DCC), a closed-form dynamical representation that enables real-time solutions for a broad class of pseudoconvex optimal control and optimization problems and demonstrates that real-time optimal control can admit a direct closed-form representation. Unlike classical formulations with Lagrange multipliers or adjoint state variables, the proposed DCC embeds constraints directly within the system dynamics. The derivation further reveals a structural equivalence between interior-point optimization and nonlinear feedback control, linking constrained optimization with classical stability theory. Through theoretical analysis and real-world-relevant numerical studies - including autonomous system control, biomechanics monitoring, and economic decision processes - we show that DCC achieves accurate optimal behavior even for highly nonlinear and high-dimensional systems. The numerical benchmark experiments suggest that DCC controls a 1000-dimensional system at 1 kHz using 77% of a modern CPU, whereas sequential quadratic programming, a widely used state-of-the-art solver for this class of problems, requires more than 80 processors to achieve comparable performance. Beyond its computational advantages, DCC provides a system-level, causally deterministic interpretation of constrained optimization, revealing that optimal behavior in a broad class of optimal control problems can emerge as the stable evolution of the system itself. These results lay the foundation for extending this perspective to more general problem classes.