Preference-Based RL

Recent advances in large video-language models (VLMs) rely on extensive fine-tuning techniques that strengthen alignment between textual and visual comprehension. Leading pipelines typically pair supervised fine-tuning (SFT) with reinforcement learning from preference data to enhance video comprehension. However, as VLMs scale in parameter size, so does the cost of gathering enough human feedback. To make fine-tuning more cost-effective, recent frameworks explore reinforcement learning with AI feedback (RLAIF), which replace human preference with AI as a judge. Current RLAIF frameworks rely on a specialized reward model trained with video narratives to create calibrated scalar rewards -- an expensive and restrictive pipeline. We propose Oracle-RLAIF, a novel framework that replaces the trained reward model with a more general Oracle ranker which acts as a drop-in model ranking candidate model responses rather than scoring them. Alongside Oracle-RLAIF, we introduce $GRPO_{rank}$, a novel rank-based loss function based on Group Relative Policy Optimization (GRPO) that directly optimizes ordinal feedback with rank-aware advantages. Empirically, we demonstrate that Oracle-RLAIF consistently outperforms leading VLMs using existing fine-tuning methods when evaluated across various video comprehension benchmarks. Oracle-RLAIF paves the path to creating flexible and data-efficient frameworks for aligning large multi-modal video models with reinforcement learning from rank rather than score.

Preference-based reinforcement learning (PbRL) enables agents to learn control policies without requiring manually designed reward functions, making it well-suited for tasks where objectives are difficult to formalize or inherently subjective. Acrobatic flight poses a particularly challenging problem due to its complex dynamics, rapid movements, and the importance of precise execution. However, manually designed reward functions for such tasks often fail to capture the qualities that matter: we find that hand-crafted rewards agree with human judgment only 60.7% of the time, underscoring the need for preference-driven approaches. In this work, we propose Reward Ensemble under Confidence (REC), a probabilistic reward learning framework for PbRL that explicitly models per-timestep reward uncertainty through an ensemble of distributional reward models. By propagating uncertainty into the preference loss and leveraging disagreement for exploration, REC achieves 88.4% of shaped reward performance on acrobatic quadrotor control, compared to 55.2% with standard Preference PPO. We train policies in simulation and successfully transfer them zero-shot to the real world, demonstrating complex acrobatic maneuvers learned purely from preference feedback. We further validate REC on a continuous control benchmark, confirming its applicability beyond the domain of aerial robotics.

In dueling bandits, the learner compares two arms in each round and observes only which one is preferred. This paper asks what changes when the compared pair also affects the next state, as in a recommendation interface where the pair shown to a user affects the user's next context. We formulate this problem as a state-wise preference-based MDP, an episodic tabular MDP in which the learner selects a pair of arms at each visited state, observes only a noisy comparison between them, and moves to a next state that may depend on the pair. The preferences may change adversarially across episodes, and the loss of a pair is defined through Borda scores. We show that the minimax regret of this problem under known transitions is $\tildeΘ(\min\{HT,(H^3SK)^{1/3}T^{2/3}\})$, where $H$, $S$, $K$, and $T$ are the numbers of steps per episode, states, arms, and episodes, and that this rate is achieved by an algorithm based on online linear optimization over the set of occupancy measures. If the pair did not affect the next state, each state would be a separate dueling bandit, and the $T^{2/3}$ term of the minimax regret would be smaller by a factor of $H^{1/3}$, demonstrating the cost of having action-dependent state transitions. We also give a policy optimization algorithm with closed-form updates, whose regret is $\tilde{O}((H^4SK^3)^{1/3}T^{2/3})$ up to lower-order terms, and we extend both algorithms to unknown transitions.

Preference-based Reinforcement Learning (PbRL) entails a variety of approaches for aligning models with human intent to alleviate the burden of reward engineering. However, most previous PbRL work has not investigated the robustness to labeler errors, inevitable with labelers who are non-experts or operate under time constraints. We introduce Similarity as Reward Alignment (SARA), a simple contrastive framework that is both resilient to noisy labels and adaptable to diverse feedback formats. SARA learns a latent representation of preferred samples and computes rewards as similarities to the learned latent. On preference data with varying realistic noise rates, we demonstrate competitive and more stable performance on continuous control offline RL benchmarks, with statistically significant improvements over baselines (Wilcoxon signed-rank, p < 0.01). We also compute correlation to the environment rewards as a proxy for measuring alignment to the underlying preference criteria. We show that the SARA computed rewards display higher correlation across noise rates compared to baselines.

This paper studies reinforcement learning from human feedback (RLHF) for aligning large language models with human preferences. While RLHF has demonstrated promising results, many algorithms are highly sensitive to misspecifications in the underlying preference model (e.g., the Bradley-Terry model), the reference policy, or the reward function, resulting in undesirable fine-tuning. To address model misspecification, we propose a doubly robust preference optimization algorithm that remains consistent when either the preference model or the reference policy is correctly specified (without requiring both). Our proposal demonstrates superior and more robust performance than state-of-the-art algorithms, both in theory and in practice. The code is available at https://github.com/DRPO4LLM/DRPO4LLM

Reward models (RMs) are central to aligning large language models (LLMs) with human values, yet noisy preference feedback can induce reward misgeneralization through spurious correlations and overfitting. We characterize the training dynamics of preference pairs and uncover a distinctive optimization signature of noise: persistently high losses, sharp temporal fluctuations, and irregular gradient updates. Our analyses further show that excluding such pairs improves downstream alignment. In practice, however, reliable preferences are not known a priori, and self-filtering based on a model's own training signals can amplify confirmation bias. To address this challenge, we introduce Collaborative Reward Modeling (CRM), an online framework for identifying reliable preferences. CRM couples cross-model peer review with curriculum learning: two RMs evaluate each other's sample selections, while curriculum keeps their learning progress synchronized. Across extensive experiments, CRM improves RM generalization by up to 9.94 points on RewardBench under 40% preference noise. CRM further generalizes to implicit-reward alignment, offering a robust and broadly applicable approach to learning from imperfect feedback.

We study data corruption robustness for reinforcement learning with human feedback (RLHF) in an offline setting. Given an offline dataset of pairs of trajectories along with feedback about human preferences, an $\varepsilon$-fraction of the pairs is corrupted (e.g., feedback flipped or trajectory features manipulated), capturing an adversarial attack or noisy human preferences. We aim to design algorithms that identify a near-optimal policy from the corrupted data, with provable guarantees. Existing theoretical works have separately studied the settings of corruption robust RL (learning from scalar rewards directly under corruption) and offline RLHF (learning from human feedback without corruption); however, they are inapplicable to our problem of dealing with corrupted data in offline RLHF setting. To this end, we design novel corruption robust offline RLHF methods under various assumptions on the coverage of the data-generating distributions. At a high level, our methodology robustifies an offline RLHF framework by first learning a reward model along with confidence sets and then learning a pessimistic optimal policy over the confidence set. Our key insight is that learning optimal policy can be done by leveraging an offline corruption-robust RL oracle in different ways (e.g., zero-order oracle or first-order oracle), depending on the data coverage assumptions. To our knowledge, ours is the first work that provides provable corruption robust offline RLHF methods.