steplevel

In this tutorial, we explore Online Process Reward Learning (OPRL) and demonstrate how we can learn dense, step-level reward signals from trajectory preferences to solve sparse-reward reinforcement learning tasks. We …