Pith. sign in

REVIEW 7 cited by

Finetuning from Offline Reinforcement Learning: Challenges, Trade-offs and Practical Solutions

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2303.17396 v1 pith:KHA2LDXT submitted 2023-03-30 cs.LG

Finetuning from Offline Reinforcement Learning: Challenges, Trade-offs and Practical Solutions

classification cs.LG
keywords offlineonlinelearningpolicyfinetuningperformanceagentsalgorithms
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Offline reinforcement learning (RL) allows for the training of competent agents from offline datasets without any interaction with the environment. Online finetuning of such offline models can further improve performance. But how should we ideally finetune agents obtained from offline RL training? While offline RL algorithms can in principle be used for finetuning, in practice, their online performance improves slowly. In contrast, we show that it is possible to use standard online off-policy algorithms for faster improvement. However, we find this approach may suffer from policy collapse, where the policy undergoes severe performance deterioration during initial online learning. We investigate the issue of policy collapse and how it relates to data diversity, algorithm choices and online replay distribution. Based on these insights, we propose a conservative policy optimization procedure that can achieve stable and sample-efficient online learning from offline pretraining.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Improving Robotic Generalist Policies via Flow Reversal Steering

    cs.RO 2026-06 unverdicted novelty 7.0

    Flow Reversal Steering steers flow matching generalist policies by reversing suboptimal actions to nearby better modes, enabling improved zero-shot control, quick distillation, and RL bootstrapping in robotic manipulation.

  2. Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling

    cs.LG 2025-07 unverdicted novelty 7.0

    Prefix-RFT blends SFT and RFT via prefix sampling from demonstrations to outperform standalone SFT, RFT, and mixed-policy baselines on math reasoning problems.

  3. Evaluating Fuzz Testing for Reinforcement Learning Agents

    cs.LG 2026-07 accept novelty 6.0

    Under unified budgets, MDPFuzz leads crash count and speed; SeqDivFuzz leads diversity; fuzz crashes improve robustness and train cross-fuzzer safety monitors.

  4. Robo-ValueRL: Reliable Value Estimation for Offline-to-Online Reinforcement Learning

    cs.RO 2026-07 conditional novelty 6.0

    Reliable history-conditioned value functions enable value-guided offline-to-online RL that scales better than behavior cloning and stabilizes online adaptation on hard real-robot tasks.

  5. Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors

    cs.LG 2026-06 unverdicted novelty 6.0

    Argues for shifting to diagnosis-driven tension management of offline priors in online RL, supported by a framework on prior roles, experiments showing help-or-hurt reversals, and cross-domain evidence.

  6. Reinforcement Learning with Action Chunking

    cs.LG 2025-07 unverdicted novelty 6.0

    Q-chunking improves offline-to-online RL sample efficiency on long-horizon sparse-reward manipulation tasks by applying action chunking to TD learning.

  7. COOPO: Cyclic Offline-Online Policy Optimization Algorithm

    cs.LG 2026-05 unverdicted novelty 5.0

    COOPO is a cyclic offline-online RL algorithm that repeatedly anchors the policy to a dataset via KL-regularized updates then fine-tunes online, claiming better sample efficiency and monotonic improvement under covera...