The paper introduces TWICE (TWIn-world Credit Estimation), a method for agentic reinforcement learning that uses twin-world counterfactuals to assign credit to individual decisions within an episode. By re‑running episodes with the same sampling noise but different decisions, TWICE produces unbiased turn‑level advantages for policy‑gradient updates, reducing the number of rollouts needed compared to independent counterfactuals. Experiments on seven question‑answering benchmarks show that TWICE improves exact match scores for GRPO by 5.3 and 4.5 points on Qwen2.5‑3B and 7B, outperforms tree‑search and counterfactual‑branching baselines, and requires 68% fewer auxiliary tokens than VinePPO.
arXiv:2605.06375v2 Announce Type: replace-cross Abstract: Policy-gradient methods need a scalar reward, yet for many tasks no formula captures what we want, while judging which of two outcomes is bet...
The paper investigates the problem of sharing a single critic across multiple parallel environments in reinforcement learning. It shows that when environments assign different expected returns to the same state, a shared critic must reconcile conflicting value targets, which can distort advantage estimates and misguide policy updates. The authors propose a simple fix—providing the critic with the environment index—demonstrating through bandit models and experiments on CartPole, MuJoCo, BipedalWalker, and 16 Procgen games that this conditional critic stabilizes learning and boosts returns, achieving a 40.8% improvement in aggregate normalized return on unseen levels.
arXiv:2610.19103v1 Announce Type: new Abstract: Group-based reinforcement learning enables large language model agents to learn from sparse outcome rewards in long-horizon tasks. Recent methods deriv...
arXiv:2606. 25451v1 Announce Type: new Abstract: Estimating token-level advantages in reinforcement learning (RL) for language models remains challenging because scaling up episodic experience collection is expensive.
arXiv:2609.32791v2 Announce Type: replace Abstract: Reinforcement mid-training lets language models learn internal thoughts from unlabeled text, but efficient token-level credit assignment remains ch...
arXiv:2606. 25556v1 Announce Type: cross Abstract: Stepwise group-based RL is an attractive way to train long-horizon LLM agents without a learned critic: it reuses multiple sampled rollouts to estimate local advantages.
SkillGate is a method that trains agents to select the correct skill from a large slate during an episode by separating credit signals for skill selection and execution. It addresses the problem of selector credit starvation, where traditional outcome-rewarded RL fails to give sufficient credit to the skill-naming tokens, especially in long-horizon tasks. Experiments on five benchmarks show that SkillGate improves a 9B policy’s success rate from 40.8% to 53.2%, reduces exposure to misleading candidates, and requires fewer skill reads.
The paper introduces CASTLE, a framework for fully decentralized multi‑agent reinforcement learning that uses two offline‑trained world models to provide in‑context guidance to independent agents. A Local Dynamics World Model captures each agent’s trajectory dynamics, while a Semantic‑Social World Model predicts short‑horizon task and social outcomes for candidate actions based on counterfactual simulator rollouts. During online learning, these frozen models supply prediction logits to an independent PPO policy, enabling agents to compare potential actions prospectively rather than relying solely on scalar rewards. CASTLE outperforms existing baselines on several multi‑particle benchmark tasks, achieving the highest mean final scores across 30 seeds.
arXiv:2607. 10139v1 Announce Type: cross Abstract: Selecting the correct answer from a pool of candidate reasoning chains is the engine of test-time scaling, yet the standard selectors each carry a cost: self-consistency inherits the errors of the single model it resamples, and trained reward models need labeled data and transfer poorly off-distribution.
arXiv:2606. 29476v1 Announce Type: cross Abstract: Self-distilled agentic reinforcement learning augments trajectory-level reward with a token-level distillation loss, using as its teacher the same policy conditioned on privileged context.
Reinforcement learning (RL) has emerged as the primary paradigm for optimizing Large Language Model (LLM) agents, yet efficient critic-free algorithms (e.g., GRPO) suffer from a fundamental uniform cr...
arXiv:2608. 13564v1 Announce Type: new Abstract: Evaluating language-model agents at scale increasingly relies on a second language model as an automatic judge, because the gold signal, an executable environment reward, is expensive, slow, or unavailable at deployment time.