REVIEW 6 cited by
Efficient Online Reinforcement Learning Fine-Tuning Need Not Retain Offline Data
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
read the original abstract
The modern paradigm in machine learning involves pre-training on diverse data, followed by task-specific fine-tuning. In reinforcement learning (RL), this translates to learning via offline RL on a diverse historical dataset, followed by rapid online RL fine-tuning using interaction data. Most RL fine-tuning methods require continued training on offline data for stability and performance. However, this is undesirable because training on diverse offline data is slow and expensive for large datasets, and in principle, also limit the performance improvement possible because of constraints or pessimism on offline data. In this paper, we show that retaining offline data is unnecessary as long as we use a properly-designed online RL approach for fine-tuning offline RL initializations. To build this approach, we start by analyzing the role of retaining offline data in online fine-tuning. We find that continued training on offline data is mostly useful for preventing a sudden divergence in the value function at the onset of fine-tuning, caused by a distribution mismatch between the offline data and online rollouts. This divergence typically results in unlearning and forgetting the benefits of offline pre-training. Our approach, Warm-start RL (WSRL), mitigates the catastrophic forgetting of pre-trained initializations using a very simple idea. WSRL employs a warmup phase that seeds the online RL run with a very small number of rollouts from the pre-trained policy to do fast online RL. The data collected during warmup helps ``recalibrate'' the offline Q-function to the online distribution, allowing us to completely discard offline data without destabilizing the online RL fine-tuning. We show that WSRL is able to fine-tune without retaining any offline data, and is able to learn faster and attains higher performance than existing algorithms irrespective of whether they retain offline data or not.
Forward citations
Cited by 6 Pith papers
-
SpikeATac: A Multimodal Tactile Finger with Taxelized Dynamic Sensing for Dexterous Manipulation
A fingertip with 16-taxel PVDF dynamic sensing plus capacitive static sensing enables fast delicate grasping and, with RLHF fine-tuning, in-hand manipulation of fragile objects.
-
Do You Really Need to Pretrain Q-Functions for Online RL Fine-Tuning?
On six robot-manipulation tasks, offline Q-pretraining does not accelerate online RL fine-tuning from a pretrained policy, while seeding the replay buffer with rollouts from an ensemble of policies (IPE) improves fina...
-
OmniTacTune: Policy-Agnostic Real-World RL for Tactile Residual Adaptation of Visual Policies
A policy-agnostic two-stage real-world RL method learns tactile residual corrections on frozen visual policies, lifting contact-rich task success from 5–40% to 85–100% in under 80 minutes.
-
The Three Regimes of Offline-to-Online Reinforcement Learning
Offline-to-online fine-tuning works best when the chosen method preserves the stronger of the pretrained policy or the offline dataset; a three-regime taxonomy organizes these choices.
-
Value Flows
Value Flows fits the full return distribution in RL with a flow-matching critic and reweights its learning objective by estimated return variance; the central theoretical guarantee does not follow from the stated equations.
-
SimLauncher: Launching Sample-Efficient Real-world Robotic Reinforcement Learning via Simulation Pre-training
Simulation-pretrained policies, with digital-twin demos for critic bootstrapping and action proposals, cut real-world RL training time while reaching near-perfect success on three manipulation tasks.
Discussion (0). Sign in to comment.