Apr 01, 2026 Posterior Optimization with Clipped Objective for Bridging Efficiency and Stability in Generative Policy Learning