Rewarding Beliefs, Not Actions: Consistency-Guided Credit Assignment for Long-Horizon Agents

• URL: http://arxiv.org/abs/2605.20061v1
• Date: Tue, 19 May 2026 16:19:29 GMT
• Title: Rewarding Beliefs, Not Actions: Consistency-Guided Credit Assignment for Long-Horizon Agents
• Abstract summary: Reinforcement learning from verifiable rewards (RLVR) is a promising paradigm for improving large language model (LLM) agents on long-horizon interactive tasks.<n>We propose ReBel (Reward Belief), a process-level reinforcement learning algorithm that explicitly models structured belief states.<n>We evaluate ReBel on challenging long-horizon benchmarks, including ALFWorld and WebShop.
• Score: 5.917866758929418
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reinforcement learning from verifiable rewards (RLVR) is a promising paradigm for improving large language model (LLM) agents on long-horizon interactive tasks. However, in partially observable environments, incomplete observations cause agent beliefs to drift over time, while delayed rewards obscure the causal impact of intermediate decisions, exacerbating temporal credit assignment challenges. To address this, we propose ReBel (Reward Belief), a process-level reinforcement learning algorithm that explicitly models structured belief states to summarize interaction history and guide subsequent policy learning. ReBel introduces belief-consistency supervision, converting discrepancies between predicted beliefs and observed feedback into dense self-supervised signals without requiring external step-wise annotations or verifiers. It also employs belief-aware grouping to compare trajectories under similar belief states, yielding more robust and lower-variance advantage estimates. We evaluate ReBel on challenging long-horizon benchmarks, including ALFWorld and WebShop. ReBel improves task success by up to $20.4$ percentage points over the episode-level baseline GRPO and increases sample efficiency by $2.1\times$. These results suggest that belief-aware self-supervision is a promising direction for reliable long-horizon decision-making under partial observability. Code is available at: https://github.com/Fateyetian/Rebel.git.

Related papers

• AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning [58.76655851910778]
  AgentOPSD is a critic-free, recursive method for turn-level credit assignment in agentic reinforcement learning.<n>We evaluate AgentOPSD on ALFWorld, WebShop, and Search-QA using Qwen2.5 models at two scales (3B and 7B)
  arXiv  Detail & Related papers  (2026-08-06T13:00:59Z)
• Agentic Reinforcement Learning with Observation-Calibrated Self-Distillation [18.20369971970953]
  On-Policy Self-Distillation (OPSD) re-scores generated tokens under a privileged replay view to obtain dense, token-level supervision.<n>The resulting support may reflect both the privileged information contained in the replay view and score shifts induced by the replay scaffold.<n>We propose Observation-Calibrated Self-Distillation (OCSD) to resolve this confounding.
  arXiv  Detail & Related papers  (2026-08-05T12:52:14Z)
• TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents [51.523913302114266]
  TRACE (Turn-level Reward Assignment via Credit Estimation) is a dense credit-assignment method for agentic reinforcement learning.<n>It represents rollouts as state transitions at tool-call boundaries, obtains gold-answer log-probabilities from a frozen reference model, transforms them into log-ratio state values, and derives per-action rewards as Temporal-Difference changes in those values.<n>It substantially improves base-model tool-use ability using pure RL, without a cold-start supervised fine-tuning stage, an agentic mid-training stage, or training on live-web data.
  arXiv  Detail & Related papers  (2026-07-15T16:16:42Z)
• TRACE: A Unified Rollout Budget Allocation Framework for Efficient Agentic Reinforcement Learning [52.67756371231985]
  Reinforcement learning with verifiable rewards (RLVR) is a promising approach for enhancing reasoning and agentic behavior in large language models.<n>We introduce Tree Rollout Allocation for Contrastive Exploration (TRACE), a unified rollout allocation framework.<n>Technically, TRACE allocates rollout budget to both prompt roots and intermediate prefixes that are most likely to yield mixed terminal rewards.
  arXiv  Detail & Related papers  (2026-06-09T17:16:03Z)
• Verifiable Process Rewards for Agentic Reasoning [21.195739597726142]
  Reinforcement learning from verifiable rewards (RLVR) has improved the reasoning abilities of large language models (LLMs)<n>Most existing approaches rely on sparse outcome-level feedback.<n>We propose Verifiable Process Rewards (VPR), a framework that converts such oracles into dense turn-level supervision for reinforcement learning.
  arXiv  Detail & Related papers  (2026-05-11T10:30:53Z)
• Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Execution [79.98699884805636]
  Reasoning Execution by Multiple Listeners (REMUL) is a multi-party reinforcement learning approach.<n>REMUL builds on the hypothesis that reasoning traces which other parties can follow will be more faithful.<n>Speakers are rewarded for producing reasoning that is clear to listeners.
  arXiv  Detail & Related papers  (2026-02-18T02:55:55Z)
• FaithRL: Learning to Reason Faithfully through Step-Level Faithfulness Maximization [35.23601691819328]
  Reinforcement Learning with Verifiable Rewards (RLVR) has markedly improved the performance of Large Language Models (LLMs)<n>Most RLVR pipelines rely on sparse outcome-based rewards, providing little supervision over intermediate steps.<n>We propose FaithRL, a general reinforcement learning framework that directly optimize reasoning faithfulness.
  arXiv  Detail & Related papers  (2026-02-03T13:28:17Z)
• From Verifiable Dot to Reward Chain: Harnessing Verifiable Reference-based Rewards for Reinforcement Learning of Open-ended Generation [52.62655622099456]
  We propose reinforcement learning with verifiable reference-based rewards (RLVRR)<n>Instead of checking the final answer, RLVRR extracts an ordered linguistic signal from high-quality references (i.e., reward chain)<n>In this way, RLVRR decomposes rewards into two dimensions: content, which preserves deterministic core concepts, and style, which evaluates adherence to stylistic properties.
  arXiv  Detail & Related papers  (2026-01-26T14:39:58Z)
• Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency [78.91846841708586]
  We show that even facts answered with perfect self-consistency can rapidly collapse under mild contextual interference.<n>We propose Neighbor-Consistency Belief (NCB), a structural measure of belief that evaluates response coherence across a conceptual neighborhood.<n>We also present Structure-Aware Training (SAT), which optimize context-invariant belief structure and reduces long-tail knowledge brittleness by approximately 30%.
  arXiv  Detail & Related papers  (2026-01-09T16:23:21Z)
• Hybrid Reinforcement: When Reward Is Sparse, It's Better to Be Dense [36.71358559780692]
  HERO is a reinforcement learning framework that integrates verifier signals with reward-model scores in a structured way.<n> HERO consistently outperforms RM-only and verifier-only baselines, with strong gains on both verifiable and hard-to-verify tasks.
  arXiv  Detail & Related papers  (2025-10-08T17:09:41Z)
• Intra-Trajectory Consistency for Reward Modeling [67.84522106537274]
  We develop an intra-trajectory consistency regularization to enforce that adjacent processes with higher next-token generation probability maintain more consistent rewards.<n>We show that the reward model trained with the proposed regularization induces better DPO-aligned policies and achieves better best-of-N (BON) inference-time verification results.
  arXiv  Detail & Related papers  (2025-06-10T12:59:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.