Stochastic Optimal Control
Overview
The field seeks policies that optimize expected cost or reward for systems evolving under noise, partial information, jumps, or uncertain models, from robots and portfolios to generative models. Classical approaches use Hamilton–Jacobi–Bellman equations, dynamic programming, and Pontryagin’s maximum principle; newer methods include reinforcement learning, operator learning, model predictive control, and Schrödinger bridges. A central practical challenge is scaling these methods to high-dimensional or long-horizon problems while preserving safety, robustness, and reliable performance when models or data are imperfect. Current work balances computationally efficient learning and sampling methods against stronger convergence, stability, and constraint guarantees.
Latest papers
102 papers15 papers in the last four weeks, up 25% on the four weeks before. 0.1% of all new papers.