{"id":"arxiv/2608.23178","url":"https://arxiv.org/abs/2608.23178","tags":["math.OC","eess.SY"],"score":0,"title":"Zeroth-Order Nonsmooth Nonconvex Optimization with Convex Liftings and Its Application to State-Feedback $H_\\infty$ Policy Optimization","domain":"math.OC","source":"arxiv","authors":["Xuhao Wang","Yujie Tang"],"reviews":0,"subject":"math.OC","summary":"Direct policy optimization is widely used in reinforcement learning and control, but generally leads to nonconvex optimization problems. For state-feedback $H_\\infty$ control, the policy objective is also nonsmooth, despite possessing a benign landscape whose hidden convexity can be revealed by the recently developed extended convex lifting framework. Motivated by recent advances in hidden convex optimization, we study zeroth-order optimization of nonsmooth, nonconvex problems admitting a convex lifting. We propose a zeroth-order proximal point algorithm: An inexact proximal-point outer loop…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-24","reactions":{}}