Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

ROAD: Adaptive Data Mixing for Offline-to-Online Reinforcement Learning via Bi-Level Optimization

About

Offline-to-online reinforcement learning harnesses the stability of offline pretraining and the flexibility of online fine-tuning. A key challenge lies in the non-stationary distribution shift between offline datasets and the evolving online policy. Common approaches often rely on static mixing ratios or heuristic-based replay strategies, which lack adaptability to different environments and varying training dynamics, resulting in suboptimal tradeoff between stability and asymptotic performance. In this work, we propose Reinforcement Learning with Optimized Adaptive Data-mixing (ROAD), a dynamic plug-and-play framework that automates the data replay process. We identify a fundamental objective misalignment in existing approaches. To tackle this, we formulate the data selection problem as a bi-level optimization process, interpreting the data mixing strategy as a meta-decision governing the policy performance (outer-level) during online fine-tuning, while the conventional Q-learning updates operate at the inner level. To make it tractable, we propose a practical algorithm using a multi-armed bandit mechanism. This is guided by a surrogate objective approximating the bi-level gradient, which simultaneously maintains offline priors and prevents value overestimation. Our empirical results demonstrate that this approach consistently outperforms existing data replay methods across various datasets, eliminating the need for manual, context-specific adjustments while achieving superior stability and asymptotic performance.

Letian Yang, Xu Liu, Yiqiang Lu, Jian Liu, Weiqiang Wang, Shuai Li (1) __INSTITUTION_6__ Shanghai Jiao Tong University, Shanghai, China, (2) Ant Group, Shanghai, China)• 2026

Related benchmarks

TaskDatasetResultRank
Offline-to-Online Reinforcement LearningD4RL AntMaze
Success Rate (Large Diverse)74.51
20
Offline Reinforcement LearningD4RL Summary v0 (Aggregate)
Average Return (All Settings)71.12
13
Reinforcement LearningHalfCheetah Random
Avg Normalized Score103.5
13
Offline-to-Online Reinforcement LearningD4RL Franka Kitchen
Partial Success Rate46.65
12
Offline-to-Online Reinforcement LearningD4RL Gym-Locomotion
HalfCheetah Return (Random)88.36
10
Offline-to-Online Reinforcement LearningD4RL Aggregate
Average Normalized Score71.95
10
Reinforcement Learninghalfcheetah medium-replay
Normalized Return96.71
7
Reinforcement Learninghalfcheetah medium--
7
Offline-to-Online Reinforcement LearningD4RL hopper--
4
Offline-to-Online Reinforcement LearningHalfCheetah D4RL Locomotion
Return (HalfCheetah Random)49.37
3
Showing 10 of 11 rows

Other info

Follow for update