SOURCE-LINKED INTELLIGENCE

Bellman Policy Optimization

arXiv · AI, language, vision and robotics · article · Sep 14, 2026 · UTC

Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities of large language models (LLMs). We introduce Bellman Policy Optimization (BPO), a critic-free method derived from Policy Mirror Descent (PMD). For autoregressive generation with terminal rewards, BPO uses the Bellman equations to reformulate PMD as a trajectory-level objective. The reformulation avoids estimating state values at intermediate states. We prove that it has the same unique optimal solution as the original PMD objective. We derive the practical BPO loss by approximating this objective. Its mi

Read original source ↗ Open in workspace

recordType

paper

region

Global