Abstract

This paper presents a strictly convex chance-constrained stochastic control framework that accounts for uncertainty in control specifications such as reference trajectories and operational constraints. By jointly optimizing control inputs and risk allocation under general (possibly non-Gaussian) uncertainties, the proposed method guarantees probabilistic constraint satisfaction while ensuring strict convexity, leading to uniqueness and continuity of the optimal solution. The formulation is further extended to nonlinear model-based control using exactly linearizable models identified through machine learning. The effectiveness of the proposed approach is demonstrated through model predictive control applied to a hybrid powertrain system.

 

Quick Read (beta)