HINT-SD: Targeted Hindsight Self-Distillation for Long-Horizon Agents

• URL: http://arxiv.org/abs/2605.17873v1
• Date: Mon, 18 May 2026 05:34:03 GMT
• Title: HINT-SD: Targeted Hindsight Self-Distillation for Long-Horizon Agents
• Abstract summary: Training long-horizon LLM agents with reinforcement learning is challenging because sparse outcome rewards reveal whether a task succeeds, but not which intermediate actions caused the outcome or how they should be corrected.<n>Recent methods alleviate this issue by generating rewards or textual hints from turn-level action-output signals, or by using feedback-conditioned self-distillation.<n>We propose HINT-SD, a targeted self-distillation framework that uses full-trajectory hindsight to select failure-relevant actions and applies feedback-conditioned distillation only on targeted action spans.
• Score: 45.00250014235802
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Training long-horizon LLM agents with reinforcement learning is challenging because sparse outcome rewards reveal whether a task succeeds, but not which intermediate actions caused the outcome or how they should be corrected. Recent methods alleviate this issue by generating rewards or textual hints from turn-level action-output signals, or by using feedback-conditioned self-distillation. However, generating feedback at every turn is inefficient when many intermediate turns are already successful or neutral, and applying feedback at a fixed or misaligned turn often fails to supervise the actions that contributed to the failure. To bridge this gap, we propose HINT-SD, a targeted self-distillation framework that uses full-trajectory hindsight to select failure-relevant actions and applies feedback-conditioned distillation only on targeted action spans. Experiments on BFCL v3 and AppWorld show that our method improves over the dense per-turn feedback baseline by up to 18.80 percent while achieving 2.26$\times$ lower time per training step, suggesting that selecting where to distill is a key factor for both effective and efficient long-horizon agent training.

Related papers

• H$^2$SD: Hybrid Hindsight Self-Distillation [61.71131241473028]
  Reinforcement learning with verifiable rewards (RLVR) provides reliable outcome supervision for language model reasoning.<n>Existing self-distillation methods add a privileged teacher but typically assign it a fixed role.<n>We introduce Hybrid Hindsight Self-Distillation ($mathrmH2mathrmSD$), which jointly adapts teacher context and update strategy to trajectory correctness.
  arXiv  Detail & Related papers  (2026-07-21T10:47:27Z)
• Hierarchical Advantage Weighting for Online RL Fine-Tuning of VLAs from Sparse Episode Outcomes [40.20394793623452]
  When pretrained VLA policies are fine-tuned through online RL, each rollout episode produces only a single binary outcome.<n>We propose Hierarchical Advantage-Weighted Behavior Cloning (HABC), which trains separate critic heads for viability and efficiency.<n>HABC raises success from supervised fine-tuning (SFT) baselines of 36%, 44%, and 12% to 92%, 88%, and 38%.
  arXiv  Detail & Related papers  (2026-06-15T17:57:14Z)
• Localizing Credit at the Divergence: Path-Conditioned Self-Distillation for LLM Reasoning [12.363923974156636]
  Reinforcement learning from verifiable rewards assigns a single scalar to each rollout.<n>On-policy self-distillation addresses this by letting the same model act as a teacher conditioned on privileged information.<n>We propose Hindsight Self-Distillation (HSD), which conditions the teacher on a successful peer rollout.
  arXiv  Detail & Related papers  (2026-06-14T03:37:27Z)
• HERO: Hindsight-Enhanced Reflection from Environment Observations for Agentic Self-Distillation [50.53459634301361]
  HERO is a hindsight-enhanced self-distillation framework that uses next environment observations as locally aligned feedback.<n> HERO improves task success and reduces unnecessary turns over environment-feedback-only self-distillation and GRPO.
  arXiv  Detail & Related papers  (2026-06-10T01:35:34Z)
• Are Full Rollouts Necessary for On-Policy Distillation? [63.18243901591995]
  On-policy distillation (OPD) provides dense teacher feedback along student-generated rollouts.<n>We identify the rollout horizon as a key bottleneck in OPD that substantially impacts training efficiency.<n>We propose two simple horizon-control strategies: Progressive OPD, which gradually expands the rollout horizon during training, and Truncated OPD, which permanently performs distillation on reliable truncated rollouts.
  arXiv  Detail & Related papers  (2026-05-29T16:12:54Z)
• What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents [70.6980022118038]
  Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignment remains challenging.<n>Existing methods rely on trajectory-level rewards or proxy signals, without fully leveraging per-step environmental feedback.<n>We introduce SERL, a selective environment-reweighted learning framework.
  arXiv  Detail & Related papers  (2026-05-19T07:00:55Z)
• Learning with Rare Success but Rich Feedback via Reflection-Enhanced Self-Distillation [71.16738415436458]
  We introduce Reflection-Enhanced Self-Distillation (RESD), a framework that transforms raw failure feedback into an active source of corrective supervision.<n>RESD interprets failed trajectories by generating retrospective reflections to diagnose local errors, and curates a persistent global playbook.<n> Empirical evaluations on multiple continual learning tasks demonstrate that RESD substantially outperforms standard self-distillation baselines.
  arXiv  Detail & Related papers  (2026-05-12T20:46:05Z)
• RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models [90.39703013636868]
  RePO-VLA is a recovery-driven policy optimization framework.<n>It assigns distinct roles to success, recovery, and failure trajectories.<n>It raises adversarial success from 20% to 75% on average and up to 80% in scaled real-world trials.
  arXiv  Detail & Related papers  (2026-05-10T08:24:05Z)
• Not All Turns Matter: Credit Assignment for Multi-Turn Jailbreaking [39.11709072420052]
  We propose TRACE, a turn-aware credit assignment framework for reinforcement learning-based multi-turn jailbreaking.<n>For successful trajectories, TRACE estimates turn-level contributions via leave-one-turn-out semantic masking.<n>For failed ones, TRACE assigns penalties based on prompt harmfulness and semantic relevance.
  arXiv  Detail & Related papers  (2026-05-09T08:07:30Z)
• HELM: Harness-Enhanced Long-horizon Memory for Vision-Language-Action Manipulation [2.9231828959903474]
  Vision-Language-Action (VLA) models fail systematically on long-horizon manipulation tasks despite strong short-horizon performance.<n>We show that this failure is not resolved by extending context length alone in the current reactive execution setting.<n>We present HELM, a model-agnostic framework that addresses these deficiencies with three components.
  arXiv  Detail & Related papers  (2026-04-20T19:57:35Z)
• R$^3$L: Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification [44.99719889905381]
  Reinforcement learning drives recent advances in reasoning and agentic capabilities, yet current approaches struggle with both exploration and exploitation.<n>We propose R$3$L, Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive Amplification.<n>Experiments on agentic and reasoning tasks demonstrate 5% to 52% relative improvements over baselines while maintaining training stability.
  arXiv  Detail & Related papers  (2026-01-07T09:04:52Z)
• FlowSteer: Guiding Few-Step Image Synthesis with Authentic Trajectories [82.90132015584359]
  ReFlow has theoretical consistency with flow matching but suboptimal performance in practical scenarios.<n>We propose FlowSteer, a method unlocks the potential of ReFlow-based distillation by guiding the student along teacher's authentic generation trajectories.
  arXiv  Detail & Related papers  (2025-11-24T07:13:23Z)
• Rethinking Reward Miscalibration of GRPO in Agentic RL [18.495499496405635]
  We show that outcome based reward ensures expected negative advantage for those flawed middle steps.<n>We propose training the actor to classify good or bad actions to separate the embedding of good/bad actions.
  arXiv  Detail & Related papers  (2025-09-28T13:24:38Z)
• Augmenting Unsupervised Reinforcement Learning with Self-Reference [63.68018737038331]
  Humans possess the ability to draw on past experiences explicitly when learning new tasks. We propose the Self-Reference (SR) approach, an add-on module explicitly designed to leverage historical information. Our approach achieves state-of-the-art results in terms of Interquartile Mean (IQM) performance and Optimality Gap reduction on the Unsupervised Reinforcement Learning Benchmark.
  arXiv  Detail & Related papers  (2023-11-16T09:07:34Z)
• Correcting Momentum in Temporal Difference Learning [95.62766731469671]
  We argue that momentum in Temporal Difference (TD) learning accumulates gradients that become doubly stale. We show that this phenomenon exists, and then propose a first-order correction term to momentum. An important insight of this work is that deep RL methods are not always best served by directly importing techniques from the supervised setting.
  arXiv  Detail & Related papers  (2021-06-07T20:41:15Z)

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.