Offline-To-Online RL

Several approaches have been proposed to improve the sample efficiency of online reinforcement learning (RL) by leveraging demonstrations collected offline. The offline data can be used directly as transitions to optimize RL objectives, or offline policy and value functions can first be learned from the data and then used for online finetuning or to provide reference actions. While each of these strategies has shown compelling results, it is unclear which method has the most impact on sample efficiency, whether these approaches can be combined, and if there are cumulative benefits. We classify existing demonstration-augmented RL approaches into three categories and perform an extensive empirical study of their strengths, weaknesses, and combinations to isolate the contribution of each strategy and determine effective hybrid combinations for sample-efficient online RL. Our analysis reveals that directly reusing offline data and initializing with behavior cloning consistently outperform more complex offline RL pretraining methods for improving online sample efficiency.

While offline reinforcement learning provides reliable policies for real-world deployment, its inherent pessimism severely restricts an agent's ability to explore and collect novel data online. Drawing inspiration from safe reinforcement learning, exploring near the boundary of regions well covered by the offline dataset and reliably modeled by the simulator allows an agent to take manageable risks--venturing into informative but moderate-uncertainty states while remaining close enough to familiar regions for safe recovery. However, naively rewarding this boundary-seeking behavior can lead to a degenerate parking behavior, where the agent simply stops once it reaches the frontier. To solve this, we propose a novel vector-field reward shaping paradigm designed to induce continuous, safe boundary exploration for non-adaptive deployed policies. Operating on an uncertainty oracle trained from offline data, our reward combines two complementary components: a gradient-alignment term that attracts the agent toward a target uncertainty level, and a rotational-flow term that promotes motion along the local tangent plane of the uncertainty manifold. Through theoretical analysis, we show that this reward structure naturally induces sustained exploratory behavior along the boundary while preventing degenerate solutions. Empirically, by integrating our proposed reward shaping with Soft Actor-Critic on a 2D continuous navigation task, we validate that agents successfully traverse uncertainty boundaries while balancing safe, informative data collection with primary task completion.

We introduce Distribution Contractive Reinforcement Learning (DICE-RL), a framework that uses reinforcement learning (RL) as a "distribution contraction" operator to refine pretrained generative robot policies. DICE-RL turns a pretrained behavior prior into a high-performing "pro" policy by amplifying high-success behaviors from online feedback. We pretrain a diffusion- or flow-based policy for broad behavioral coverage, then finetune it with a stable, sample-efficient residual off-policy RL framework that combines selective behavior regularization with value-guided action selection. Extensive experiments and analyses show that DICE-RL reliably improves performance with strong stability and sample efficiency. It enables mastery of complex long-horizon manipulation skills directly from high-dimensional pixel inputs, both in simulation and on a real robot. Project website: https://zhanyisun.github.io/dice.rl.2026/.

Offline-to-online reinforcement learning (RL) offers a promising paradigm for robotics by pre-training policies on safe, offline demonstrations and fine-tuning them via online interaction. However, a fundamental challenge remains: how to safely explore online without deviating from the behavioral support of the offline data? While recent methods leverage conditional variational autoencoders (CVAEs) to bound exploration within a latent space, they inherently suffer from an exploitation gap -- a performance ceiling imposed by the decoder's reconstruction loss. We introduce SPAARS, a curriculum learning framework that initially constrains exploration to the low-dimensional latent manifold for sample-efficient, safe behavioral improvement, then seamlessly transfers control to the raw action space, bypassing the decoder bottleneck. SPAARS has two instantiations: the CVAE-based variant requires only unordered (s,a) pairs and no trajectory segmentation; SPAARS-SUPE pairs SPAARS with OPAL temporal skill pretraining for stronger exploration structure at the cost of requiring trajectory chunks. We prove an upper bound on the exploitation gap using the Performance Difference Lemma, establish that latent-space policy gradients achieve provable variance reduction over raw-space exploration, and show that concurrent behavioral cloning during the latent phase directly controls curriculum transition stability. Empirically, SPAARS-SUPE achieves 0.825 normalized return on kitchen-mixed-v0 versus 0.75 for SUPE, with 5x better sample efficiency; standalone SPAARS achieves 92.7 and 102.9 normalized return on hopper-medium-v2 and walker2d-medium-v2 respectively, surpassing IQL baselines of 66.3 and 78.3 respectively, confirming the utility of the unordered-pair CVAE instantiation.

State-of-the-art deep reinforcement learning (RL) methods have achieved remarkable performance in continuous control tasks, yet their computational complexity is often incompatible with the constraints of resource-limited hardware, due to their reliance on replay buffers, batch updates, and target networks. The emerging paradigm of streaming deep RL addresses this limitation through purely online updates, achieving strong empirical performance on standard benchmarks. In this work, we propose two novel streaming deep RL algorithms, Streaming Soft Actor-Critic (S2AC) and Streaming Deterministic Actor-Critic (SDAC), explicitly designed to be compatible with state-of-the-art batch RL methods, making them particularly suitable for on-device finetuning applications such as Sim2Real transfer. Both algorithms achieve performance comparable to state-of-the-art streaming baselines on standard benchmarks without requiring tedious per-environment hyperparameter tuning. We further investigate the batch-to-streaming transition, showing that a naive transition does not guarantee preservation of pre-trained policy performance, and propose a principled approach to address this challenge.

Precise grasp force regulation in tendon-driven surgical instruments is fundamentally limited by nonlinear coupling between motor dynamics, transmission compliance, friction, and distal mechanics. Existing solutions typically rely on distal force sensing or analytical compensation, increasing hardware complexity or degrading performance under dynamic motion. We present a sensorless control framework that combines physics-consistent modeling and hybrid reinforcement learning to achieve high-precision distal force regulation in a proximally actuated surgical end-effector. We develop a first-principles digital twin of the da Vinci Xi grasping mechanism that captures coupled electrical, transmission, and jaw dynamics within a unified differential-algebraic formulation. To safely learn control policies in this stiff and highly nonlinear system, we introduce a three-stage pipeline:(i)a receding-horizon CMA-ES oracle that generates dynamically feasible expert trajectories,(ii)fully offline policy learning via Implicit Q-Learning to ensure stable initialization without unsafe exploration, and (iii)online refinement using TD3 for adaptation to on-policy dynamics. The resulting policy directly maps proximal measurements to motor voltages and requires no distal sensing. In simulation, the controller maintains grasp force within 1% of the desired reference during multi-harmonic jaw motion. Hardware experiments demonstrate average force errors below 4% across diverse trajectories, validating sim-to-real transfer. The learned policy contains approximately 71k param and executes at kH rates, enabling real-time deployment. These results demonstrate that high-fidelity modeling combined with structured offline-online RL can recover precise distal force behavior without additional sensing, offering a scalable and mechanically compatible solution for surgical robotic manipulation.

Generative models have recently demonstrated remarkable success across diverse domains, motivating their adoption as expressive policies in reinforcement learning (RL). While they have shown strong performance in offline RL, particularly where the target distribution is well defined, their extension to online fine-tuning has largely been treated as a direct continuation of offline pre-training, leaving key challenges unaddressed. In this paper, we propose Flow Matching with Injected Noise for Offline-to-Online RL (FINO), a novel method that leverages flow matching-based policies to enhance sample efficiency for offline-to-online RL. FINO facilitates effective exploration by injecting noise into policy training, thereby encouraging a broader range of actions beyond those observed in the offline dataset. In addition to exploration-enhanced flow policy training, we combine an entropy-guided sampling mechanism to balance exploration and exploitation, allowing the policy to adapt its behavior throughout online fine-tuning. Experiments across diverse, challenging tasks demonstrate that FINO consistently achieves superior performance under limited online budgets.

Modern offline Reinforcement Learning (RL) methods find performant actor-critics, however, fine-tuning these actor-critics online with value-based RL algorithms typically causes immediate drops in performance. We provide evidence consistent with the hypothesis that, in the loss landscape, offline maxima for prior algorithms and online maxima are separated by low-performance valleys that gradient-based fine-tuning traverses. Following this, we present Score Matched Actor-Critic (SMAC), an offline RL method designed to learn actor-critics that transition to online value-based RL algorithms with no drop in performance. SMAC avoids valleys between offline and online maxima by regularizing the Q-function during the offline phase to respect a first-order derivative equality between the score of the policy and action-gradient of the Q-function. We experimentally demonstrate that SMAC converges to offline maxima that are connected to better online maxima via paths with monotonically increasing reward found by first-order optimization. SMAC achieves smooth transfer to Soft Actor-Critic and TD3 in 6/6 D4RL tasks. In 4/6 environments, it reduces regret by 34-58% over the best baseline.

Offline-to-online reinforcement learning (RL) has emerged as a practical paradigm that leverages offline datasets for pretraining and online interactions for fine-tuning. However, its empirical behavior is highly inconsistent: design choices of online fine-tuning that work well in one setting can fail completely in another. Guided by the stability--plasticity principle, we propose a framework that can explain this inconsistency: We argue that efficient fine-tuning must preserve the utility of the stronger offline prior, whether that is the pretrained policy or the offline dataset, while maintaining sufficient plasticity. This perspective identifies three regimes of online fine-tuning, each requiring distinct stability properties. We validate this framework through a large-scale empirical study, finding that the results strongly align with its predictions in 45 out of 63 cases, with only 3 opposite mismatches. This work provides a framework for guiding design choices in offline-to-online RL based on the relative performance of the offline dataset and the pretrained policy.