arXiv Machine Learning
Where to Start, How to Move: Analyzing Curricula under GRPO for Long-Horizon LLM Agents
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
Related stories
arXiv Machine Learning
Oct 15
Divergent Paths: Learning an Outcome-Predictive Representation of LLM Agent Trajectories
arXiv:2610.19614v1 Announce Type: new Abstract: In this work, we reframe agentic trajectories as paths through a shared embedding space of actions and environmental observations. We show that success...
arXiv AI
Oct 15
LLM Value Functions for Credit Assignment in Agentic RL
arXiv:2610.15840v1 Announce Type: cross Abstract: Reinforcement learning (RL) for large language model (LLM) agents remains challenging in long-horizon tasks with sparse outcome rewards, where assign...
arXiv AI
Oct 15
DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter
arXiv:2610.17666v1 Announce Type: cross Abstract: Reinforcement learning with verifiable rewards (RLVR) for multi-turn agents relies on sparse terminal rewards that say little about individual steps....
arXiv AI
Jul 24
PATS: Policy-Aware Training Scaffolding for Agentic Reinforcement Learning
arXiv:2607. 21419v1 Announce Type: new Abstract: In long-horizon LLM agent reinforcement learning, weak policies often repeat similar failures, producing uninformative rollout trajectories and limiting effective policy optimization.
arXiv AI
1d ago
AttnSR: Attention Step Reward for Long-Horizon Agentic RL
arXiv:2611.07932v1 Announce Type: cross Abstract: Long-horizon LLM agents make multi-step decisions, yet outcome-reward-only reinforcement learning provides little direct supervision for individual s...
arXiv AI
Oct 31
SCOPE: Selective Critical-Step On-Policy Self-Distillation for Agentic Reinforcement Learning
SCOPE is a new framework for selective critical‑step on‑policy self‑distillation in agentic reinforcement learning. After a supervised warm‑up, the policy examines its own trajectories, generates hindsight skills for each step, and distills guidance only at selected error steps, using a bounded teacher‑student coefficient and a success‑relative pairwise length reward. Experiments on ALFWorld and WebShop show consistent performance gains across multiple model backbones, with Qwen2.5‑3B achieving 97.7 % success on ALFWorld and 85.2 % on WebShop, and Qwen3‑32B reaching 99.2 % on seen tasks and 94.0 % on unseen tasks.