1
StepPO: Step-Aligned Policy Optimization for Agentic Reinforcement Learning
提出Step-aligned策略优化方法，解决代理强化学习中的信用分配难题，实验验证显著提升性能。
arXiv:2604.18401v2 Announce Type: replace Abstract: Agentic reinforcement learning (RL) is emerging as a critical post-training paradigm for improving…