New Meta-State-Space SMPC slashes computation cost, shapes full PDF
Efficient uncertainty propagation with unprecedented output PDF shaping capability.
Stochastic model-predictive control (SMPC) is a powerful framework for controlling stochastic dynamical systems, but existing methods often face computational challenges due to the need for accurate stochastic inference. To maintain tractability, most approaches approximate uncertainty propagation with Gaussian distributions, which can be overly conservative and limit the representation of state evolution and guarantees.
A team led by Bendegúz Györök (with Roland Tóth, Maarten Schoukens, Tamás Péni) has introduced a novel SMPC formulation based on the meta-state-space (MSS) representation of stochastic systems. This method offers a computationally efficient way to forward propagate uncertainty while retaining flexible, highly accurate approximations. Unprecedentedly, it allows direct shaping of the entire output probability density function. The preprint (arXiv:2605.26626, submitted to Automatica) includes a detailed theoretical analysis and validation via extensive simulations, showing potential for real-world applications where precise control under uncertainty is critical.
- Computationally efficient uncertainty propagation using meta-state-space representation.
- Direct shaping of the entire output probability density function, a first among SMPC methods.
- Theoretical analysis and extensive simulation study validate the approach's accuracy and flexibility.
Why It Matters
Enables more precise, efficient control of stochastic systems in robotics, autonomous vehicles, and industrial automation.