Amy Zhang

Dinghuai Zhang, Aaron Courville, Yoshua Bengio et al. · mila

While the maximum entropy (MaxEnt) reinforcement learning (RL) framework -- often touted for its exploration and robustness capabilities -- is usually motivated from a probabilistic perspective, the use of deep probabilistic models has not gained much traction in practice due to their inherent complexity. In this work, we propose the adoption of latent variable policies within the MaxEnt framework, which we show can provably approximate any policy distribution, and additionally, naturally emerges under the use of world models with a latent belief state. We discuss why latent variable policies are difficult to train, how naive approaches can fail, then subsequently introduce a series of improvements centered around low-cost marginalization of the latent state, allowing us to make full use of the latent state at minimal additional cost. We instantiate our method under the actor-critic framework, marginalizing both the actor and critic. The resulting algorithm, referred to as Stochastic Marginal Actor-Critic (SMAC), is simple yet effective. We experimentally validate our method on continuous control tasks, showing that effective marginalization can lead to better exploration and more robust training. Our implementation is open sourced at https://github.com/zdhNarsil/Stochastic-Marginal-Actor-Critic.

Tongzhou Wang, Antonio Torralba, Phillip Isola et al. · mit

In goal-reaching reinforcement learning (RL), the optimal value function has a particular geometry, called quasimetric structure. This paper introduces Quasimetric Reinforcement Learning (QRL), a new RL method that utilizes quasimetric models to learn optimal value functions. Distinct from prior approaches, the QRL objective is specifically designed for quasimetrics, and provides strong theoretical recovery guarantees. Empirically, we conduct thorough analyses on a discretized MountainCar environment, identifying properties of QRL and its advantages over alternatives. On offline and online goal-reaching benchmarks, QRL also demonstrates improved sample efficiency and performance, across both state-based and image-based observations.

Tongzhou Wang, Simon S. Du, Antonio Torralba et al. · mit

The ability to separate signal from noise, and reason with clean abstractions, is critical to intelligence. With this ability, humans can efficiently perform real world tasks without considering all possible nuisance factors.How can artificial agents do the same? What kind of information can agents safely discard as noises? In this work, we categorize information out in the wild into four types based on controllability and relation with reward, and formulate useful information as that which is both controllable and reward-relevant. This framework clarifies the kinds information removed by various prior work on representation learning in reinforcement learning (RL), and leads to our proposed approach of learning a Denoised MDP that explicitly factors out certain noise distractors. Extensive experiments on variants of DeepMind Control Suite and RoboDesk demonstrate superior performance of our denoised world model over using raw observations alone, and over prior works, across policy optimization control tasks as well as the non-control task of joint position regression.

Philippe Hansen-Estruch, Amy Zhang, Ashvin Nair et al.

Building generalizable goal-conditioned agents from rich observations is a key to reinforcement learning (RL) solving real world problems. Traditionally in goal-conditioned RL, an agent is provided with the exact goal they intend to reach. However, it is often not realistic to know the configuration of the goal before performing a task. A more scalable framework would allow us to provide the agent with an example of an analogous task, and have the agent then infer what the goal should be for its current state. We propose a new form of state abstraction called goal-conditioned bisimulation that captures functional equivariance, allowing for the reuse of skills to achieve new goals. We learn this representation using a metric form of this abstraction, and show its ability to generalize to new goals in simulation manipulation tasks. Further, we prove that this learned representation is sufficient not only for goal conditioned tasks, but is amenable to any downstream task described by a state-only reward function. Videos can be found at https://sites.google.com/view/gc-bisimulation.

Yecheng Jason Ma, Shagun Sodhani, Dinesh Jayaraman et al.

Reward and representation learning are two long-standing challenges for learning an expanding set of robot manipulation skills from sensory observations. Given the inherent cost and scarcity of in-domain, task-specific robot data, learning from large, diverse, offline human videos has emerged as a promising path towards acquiring a generally useful visual representation for control; however, how these human videos can be used for general-purpose reward learning remains an open question. We introduce $\textbf{V}$alue-$\textbf{I}$mplicit $\textbf{P}$re-training (VIP), a self-supervised pre-trained visual representation capable of generating dense and smooth reward functions for unseen robotic tasks. VIP casts representation learning from human videos as an offline goal-conditioned reinforcement learning problem and derives a self-supervised dual goal-conditioned value-function objective that does not depend on actions, enabling pre-training on unlabeled human videos. Theoretically, VIP can be understood as a novel implicit time contrastive objective that generates a temporally smooth embedding, enabling the value function to be implicitly defined via the embedding distance, which can then be used to construct the reward for any goal-image specified downstream task. Trained on large-scale Ego4D human videos and without any fine-tuning on in-domain, task-specific data, VIP's frozen representation can provide dense visual reward for an extensive set of simulated and $\textbf{real-robot}$ tasks, enabling diverse reward-based visual control methods and significantly outperforming all prior pre-trained representations. Notably, VIP can enable simple, $\textbf{few-shot}$ offline RL on a suite of real-world robot tasks with as few as 20 trajectories.

Chris Lengerich, Gabriel Synnaeve, Amy Zhang et al. · meta-ai

Traditional approaches to RL have focused on learning decision policies directly from episodic decisions, while slowly and implicitly learning the semantics of compositional representations needed for generalization. While some approaches have been adopted to refine representations via auxiliary self-supervised losses while simultaneously learning decision policies, learning compositional representations from hand-designed and context-independent self-supervised losses (multi-view) still adapts relatively slowly to the real world, which contains many non-IID subspaces requiring rapid distribution shift in both time and spatial attention patterns at varying levels of abstraction. In contrast, supervised language model cascades have shown the flexibility to adapt to many diverse manifolds, and hints of self-learning needed for autonomous task transfer. However, to date, transfer methods for language models like few-shot learning and fine-tuning still require human supervision and transfer learning using self-learning methods has been underexplored. We propose a self-supervised loss policy called contrastive distillation which manifests latent variables with high mutual information with both source and target tasks from weights to tokens. We show how this outperforms common methods of transfer learning and suggests a useful design axis of trading off compute for generalizability for online transfer. Contrastive distillation is improved through sampling from memory and suggests a simple algorithm for more efficiently sampling negative examples for contrastive losses than random sampling.

Qiyang Li, Jason Zhang, Dibya Ghosh et al. · berkeley

Learning to solve tasks from a sparse reward signal is a major challenge for standard reinforcement learning (RL) algorithms. However, in the real world, agents rarely need to solve sparse reward tasks entirely from scratch. More often, we might possess prior experience to draw on that provides considerable guidance about which actions and outcomes are possible in the world, which we can use to explore more effectively for new tasks. In this work, we study how prior data without reward labels may be used to guide and accelerate exploration for an agent solving a new sparse reward task. We propose a simple approach that learns a reward model from online experience, labels the unlabeled prior data with optimistic rewards, and then uses it concurrently alongside the online data for downstream policy and critic optimization. This general formula leads to rapid exploration in several challenging sparse-reward domains where tabula rasa exploration is insufficient, including the AntMaze domain, Adroit hand manipulation domain, and a visual simulated robotic manipulation domain. Our results highlight the ease of incorporating unlabeled prior data into existing online RL algorithms, and the (perhaps surprising) effectiveness of doing so.

Yecheng Jason Ma, William Liang, Vaidehi Som et al.

We present Language-Image Value learning (LIV), a unified objective for vision-language representation and reward learning from action-free videos with text annotations. Exploiting a novel connection between dual reinforcement learning and mutual information contrastive learning, the LIV objective trains a multi-modal representation that implicitly encodes a universal value function for tasks specified as language or image goals. We use LIV to pre-train the first control-centric vision-language representation from large human video datasets such as EpicKitchen. Given only a language or image goal, the pre-trained LIV model can assign dense rewards to each frame in videos of unseen robots or humans attempting that task in unseen environments. Further, when some target domain-specific data is available, the same objective can be used to fine-tune and improve LIV and even other pre-trained representations for robotic control and reward specification in that domain. In our experiments on several simulated and real-world robot environments, LIV models consistently outperform the best prior input state representations for imitation learning, as well as reward specification methods for policy synthesis. Our results validate the advantages of joint vision-language representation and reward learning within the unified, compact LIV framework.

Michael Chang, Alyssa L. Dayan, Franziska Meier et al.

Object rearrangement is a challenge for embodied agents because solving these tasks requires generalizing across a combinatorially large set of configurations of entities and their locations. Worse, the representations of these entities are unknown and must be inferred from sensory percepts. We present a hierarchical abstraction approach to uncover these underlying entities and achieve combinatorial generalization from unstructured visual inputs. By constructing a factorized transition graph over clusters of entity representations inferred from pixels, we show how to learn a correspondence between intervening on states of entities in the agent's model and acting on objects in the environment. We use this correspondence to develop a method for control that generalizes to different numbers and configurations of objects, which outperforms current offline deep RL methods when evaluated on simulated rearrangement tasks.

Seth Karten, Jake Grigsby, Tersoo Upaa et al.

We present the PokeAgent Challenge, a large-scale benchmark for decision-making research built on Pokemon's multi-agent battle system and expansive role-playing game (RPG) environment. Partial observability, game-theoretic reasoning, and long-horizon planning remain open problems for frontier AI, yet few benchmarks stress all three simultaneously under realistic conditions. PokeAgent targets these limitations at scale through two complementary tracks: our Battling Track, which calls for strategic reasoning and generalization under partial observability in competitive Pokemon battles, and our Speedrunning Track, which requires long-horizon planning and sequential decision-making in the Pokemon RPG. Our Battling Track supplies a dataset of 20M+ battle trajectories alongside a suite of heuristic, RL, and LLM-based baselines capable of high-level competitive play. Our Speedrunning Track provides the first standardized evaluation framework for RPG speedrunning, including an open-source multi-agent orchestration system for modular, reproducible comparisons of harness-based LLM approaches. Our NeurIPS 2025 competition validates both the quality of our resources and the research community's interest in Pokemon, with over 100 teams competing across both tracks and winning solutions detailed in our paper. Participant submissions and our baselines reveal considerable gaps between generalist (LLM), specialist (RL), and elite human performance. Analysis against the BenchPress evaluation matrix shows that Pokemon battling is nearly orthogonal to standard LLM benchmarks, measuring capabilities not captured by existing suites and positioning Pokemon as an unsolved benchmark that can drive RL and LLM research forward. We transition to a living benchmark with a live leaderboard for Battling and self-contained evaluation for Speedrunning at https://pokeagentchallenge.com.

Andrew Szot, Amy Zhang, Dhruv Batra et al.

How well do reward functions learned with inverse reinforcement learning (IRL) generalize? We illustrate that state-of-the-art IRL algorithms, which maximize a maximum-entropy objective, learn rewards that overfit to the demonstrations. Such rewards struggle to provide meaningful rewards for states not covered by the demonstrations, a major detriment when using the reward to learn policies in new situations. We introduce BC-IRL a new inverse reinforcement learning method that learns reward functions that generalize better when compared to maximum-entropy IRL approaches. In contrast to the MaxEnt framework, which learns to maximize rewards around demonstrations, BC-IRL updates reward parameters such that the policy trained with the new reward matches the expert demonstrations better. We show that BC-IRL learns rewards that generalize better on an illustrative simple task and two continuous robotic control tasks, achieving over twice the success rate of baselines in challenging generalization settings.

Martin Klissarov, Pierluca D'Oro, Shagun Sodhani et al.

Exploring rich environments and evaluating one's actions without prior knowledge is immensely challenging. In this paper, we propose Motif, a general method to interface such prior knowledge from a Large Language Model (LLM) with an agent. Motif is based on the idea of grounding LLMs for decision-making without requiring them to interact with the environment: it elicits preferences from an LLM over pairs of captions to construct an intrinsic reward, which is then used to train agents with reinforcement learning. We evaluate Motif's performance and behavior on the challenging, open-ended and procedurally-generated NetHack game. Surprisingly, by only learning to maximize its intrinsic reward, Motif achieves a higher game score than an algorithm directly trained to maximize the score itself. When combining Motif's intrinsic reward with the environment reward, our method significantly outperforms existing approaches and makes progress on tasks where no advancements have ever been made without demonstrations. Finally, we show that Motif mostly generates intuitive human-aligned behaviors which can be steered easily through prompt modifications, while scaling well with the LLM size and the amount of information given in the prompt.

Harshit Sikchi, Qinqing Zheng, Amy Zhang et al.

The goal of reinforcement learning (RL) is to find a policy that maximizes the expected cumulative return. It has been shown that this objective can be represented as an optimization problem of state-action visitation distribution under linear constraints. The dual problem of this formulation, which we refer to as dual RL, is unconstrained and easier to optimize. In this work, we first cast several state-of-the-art offline RL and offline imitation learning (IL) algorithms as instances of dual RL approaches with shared structures. Such unification allows us to identify the root cause of the shortcomings of prior methods. For offline IL, our analysis shows that prior methods are based on a restrictive coverage assumption that greatly limits their performance in practice. To fix this limitation, we propose a new discriminator-free method ReCOIL that learns to imitate from arbitrary off-policy data to obtain near-expert performance. For offline RL, our analysis frames a recent offline RL method XQL in the dual framework, and we further propose a new method f-DVL that provides alternative choices to the Gumbel regression loss that fixes the known training instability issue of XQL. The performance improvements by both of our proposed methods, ReCOIL and f-DVL, in IL and RL are validated on an extensive suite of simulated robot locomotion and manipulation tasks. Project code and details can be found at this https://hari-sikchi.github.io/dual-rl.

Andre Ye, Quan Ze Chen, Amy Zhang · uw

Medical image segmentation modeling is a high-stakes task where understanding of uncertainty is crucial for addressing visual ambiguity. Prior work has developed segmentation models utilizing probabilistic or generative mechanisms to infer uncertainty from labels where annotators draw a singular boundary. However, as these annotations cannot represent an individual annotator's uncertainty, models trained on them produce uncertainty maps that are difficult to interpret. We propose a novel segmentation representation, Confidence Contours, which uses high- and low-confidence ``contours'' to capture uncertainty directly, and develop a novel annotation system for collecting contours. We conduct an evaluation on the Lung Image Dataset Consortium (LIDC) and a synthetic dataset. From an annotation study with 30 participants, results show that Confidence Contours provide high representative capacity without considerably higher annotator effort. We also find that general-purpose segmentation models can learn Confidence Contours at the same performance level as standard singular annotations. Finally, from interviews with 5 medical experts, we find that Confidence Contour maps are more interpretable than Bayesian maps due to representation of structural uncertainty.

Aditya Mohan, Amy Zhang, Marius Lindauer

Reinforcement Learning (RL), bolstered by the expressive capabilities of Deep Neural Networks (DNNs) for function approximation, has demonstrated considerable success in numerous applications. However, its practicality in addressing various real-world scenarios, characterized by diverse and unpredictable dynamics, noisy signals, and large state and action spaces, remains limited. This limitation stems from poor data efficiency, limited generalization capabilities, a lack of safety guarantees, and the absence of interpretability, among other factors. To overcome these challenges and improve performance across these crucial metrics, one promising avenue is to incorporate additional structural information about the problem into the RL learning process. Various sub-fields of RL have proposed methods for incorporating such inductive biases. We amalgamate these diverse methodologies under a unified framework, shedding light on the role of structure in the learning problem, and classify these methods into distinct patterns of incorporating structure. By leveraging this comprehensive framework, we provide valuable insights into the challenges of structured RL and lay the groundwork for a design pattern perspective on RL research. This novel perspective paves the way for future advancements and aids in developing more effective and efficient RL algorithms that can potentially handle real-world scenarios better.

Manas Pathak, Xingyao Chen, Shuozhe Li et al.

Should we trust Large Language Models (LLMs) with high accuracy? LLMs achieve high accuracy on reasoning benchmarks, but correctness alone does not reveal the quality of the reasoning used to produce it. This highlights a fundamental limitation of outcome-based evaluation: models may arrive at correct answers through flawed reasoning, and models with substantially different reasoning capabilities can nevertheless exhibit similar benchmark accuracy, for example due to memorization or over-optimization. In this paper, we ask: given existing benchmarks, can we move beyond outcome-based evaluation to assess the quality of reasoning itself? We seek metrics that (1) differentiate models with similar accuracy and (2) are robust to variations in input prompts and generation configurations. To this end, we propose a reasoning score that evaluates reasoning traces along dimensions such as faithfulness, coherence, utility, and factuality. A remaining question is how to aggregate this score across multiple sampled traces. Naively averaging them is undesirable, particularly in long-horizon settings, where the number of possible trajectories grows rapidly, and low-confidence correct traces are more likely to be coincidental. To address this, we introduce the Filtered Reasoning Score (FRS), which computes reasoning quality using only the top-K% most confident traces. Evaluating with FRS, models that are indistinguishable under standard accuracy exhibit significant differences in reasoning quality. Moreover, models with higher FRS on one benchmark tend to perform better on other reasoning benchmarks, in both accuracy and reasoning quality. Together, these findings suggest that FRS complements accuracy by capturing a model's transferable reasoning capabilities. We open source our evaluation codebase: https://github.com/Manas2006/benchmark_reproducibility.

Edwin Zhang, Yujie Lu, Shinda Huang et al.

Training generalist agents is difficult across several axes, requiring us to deal with high-dimensional inputs (space), long horizons (time), and generalization to novel tasks. Recent advances with architectures have allowed for improved scaling along one or two of these axes, but are still computationally prohibitive to use. In this paper, we propose to address all three axes by leveraging \textbf{L}anguage to \textbf{C}ontrol \textbf{D}iffusion models as a hierarchical planner conditioned on language (LCD). We effectively and efficiently scale diffusion models for planning in extended temporal, state, and task dimensions to tackle long horizon control problems conditioned on natural language instructions, as a step towards generalist agents. Comparing LCD with other state-of-the-art models on the CALVIN language robotics benchmark finds that LCD outperforms other SOTA methods in multi-task success rates, whilst improving inference speed over other comparable diffusion models by 3.3x~15x. We show that LCD can successfully leverage the unique strength of diffusion models to produce coherent long range plans while addressing their weakness in generating low-level details and control.

Harshit Sikchi, Rohan Chitnis, Ahmed Touati et al.

Offline Goal-Conditioned Reinforcement Learning (GCRL) is tasked with learning to achieve multiple goals in an environment purely from offline datasets using sparse reward functions. Offline GCRL is pivotal for developing generalist agents capable of leveraging pre-existing datasets to learn diverse and reusable skills without hand-engineering reward functions. However, contemporary approaches to GCRL based on supervised learning and contrastive learning are often suboptimal in the offline setting. An alternative perspective on GCRL optimizes for occupancy matching, but necessitates learning a discriminator, which subsequently serves as a pseudo-reward for downstream RL. Inaccuracies in the learned discriminator can cascade, negatively influencing the resulting policy. We present a novel approach to GCRL under a new lens of mixture-distribution matching, leading to our discriminator-free method: SMORe. The key insight is combining the occupancy matching perspective of GCRL with a convex dual formulation to derive a learning objective that can better leverage suboptimal offline data. SMORe learns scores or unnormalized densities representing the importance of taking an action at a state for reaching a particular goal. SMORe is principled and our extensive experiments on the fully offline GCRL benchmark composed of robot manipulation and locomotion tasks, including high-dimensional observations, show that SMORe can outperform state-of-the-art baselines by a significant margin.

Liyuan Mao, Haoran Xu, Xianyuan Zhan et al.

One important property of DIstribution Correction Estimation (DICE) methods is that the solution is the optimal stationary distribution ratio between the optimized and data collection policy. In this work, we show that DICE-based methods can be viewed as a transformation from the behavior distribution to the optimal policy distribution. Based on this, we propose a novel approach, Diffusion-DICE, that directly performs this transformation using diffusion models. We find that the optimal policy's score function can be decomposed into two terms: the behavior policy's score function and the gradient of a guidance term which depends on the optimal distribution ratio. The first term can be obtained from a diffusion model trained on the dataset and we propose an in-sample learning objective to learn the second term. Due to the multi-modality contained in the optimal policy distribution, the transformation in Diffusion-DICE may guide towards those local-optimal modes. We thus generate a few candidate actions and carefully select from them to approach global-optimum. Different from all other diffusion-based offline RL methods, the guide-then-select paradigm in Diffusion-DICE only uses in-sample actions for training and brings minimal error exploitation in the value function. We use a didatic toycase example to show how previous diffusion-based methods fail to generate optimal actions due to leveraging these errors and how Diffusion-DICE successfully avoids that. We then conduct extensive experiments on benchmark datasets to show the strong performance of Diffusion-DICE. Project page at https://ryanxhr.github.io/Diffusion-DICE/.

Pranaya Jajoo, Harshit Sikchi, Siddhant Agarwal et al.

Behavioral Foundation Models (BFMs) produce agents with the capability to adapt to any unknown reward or task. These methods, however, are only able to produce near-optimal policies for the reward functions that are in the span of some pre-existing state features, making the choice of state features crucial to the expressivity of the BFM. As a result, BFMs are trained using a variety of complex objectives and require sufficient dataset coverage, to train task-useful spanning features. In this work, we examine the question: are these complex representation learning objectives necessary for zero-shot RL? Specifically, we revisit the objective of self-supervised next-state prediction in latent space for state feature learning, but observe that such an objective alone is prone to increasing state-feature similarity, and subsequently reducing span. We propose an approach, Regularized Latent Dynamics Prediction (RLDP), that adds a simple orthogonality regularization to maintain feature diversity and can match or surpass state-of-the-art complex representation learning methods for zero-shot RL. Furthermore, we empirically show that prior approaches perform poorly in low-coverage scenarios where RLDP still succeeds.

Hanlin Zhu, Amy Zhang

Goal-conditioned reinforcement learning (GCRL) refers to learning general-purpose skills that aim to reach diverse goals. In particular, offline GCRL only requires purely pre-collected datasets to perform training tasks without additional interactions with the environment. Although offline GCRL has become increasingly prevalent and many previous works have demonstrated its empirical success, the theoretical understanding of efficient offline GCRL algorithms is not well established, especially when the state space is huge and the offline dataset only covers the policy we aim to learn. In this paper, we provide a rigorous theoretical analysis of an existing empirically successful offline GCRL algorithm. We prove that under slight modification, this algorithm enjoys an $\widetilde{O}(\text{poly}(1/ε))$ sample complexity (where $ε$ is the desired suboptimality of the learned policy) with general function approximation thanks to the property of (semi-)strong convexity of the objective functions. We only require nearly minimal assumptions on the dataset (single-policy concentrability) and the function class (realizability). Moreover, this algorithm consists of two uninterleaved optimization steps, which we refer to as $V$-learning and policy learning, and is computationally stable since it does not involve minimax optimization. We also empirically validate our theory by showing that the modified algorithm outperforms the previous algorithm in various real-world environments. To the best of our knowledge, this is the first algorithm that is both provably efficient with general function approximation and single-policy concentrability, and empirically successful without requiring solving minimax optimization problems.

Sarthak Dayal, Abhinav Peri, Carl Qi et al.

Hierarchical Reinforcement Learning (HRL) promises to solve long-horizon Reinforcement Learning (RL) tasks more efficiently than non-hierarchical counterparts by discovering and reusing temporally-extended skills. However, obtaining skills that are actually reusable remains an open challenge. Towards this end, we focus on abstractions that exploit the intuition of local dynamics: local transitions in different global contexts require similar kinds of action sequences. By aligning these contexts with the action sequences they require, we are able to learn which skills to reuse and where to reuse them. In principle, this information should benefit many HRL algorithms, where high-level policies have to reason about the low-level skills they use. The resulting algorithm CARL (Contrastive Action-based Representations for Reusable Local Control) shows both qualitative clustering of meaningful skills in complex humanoid environments and improved downstream performance on the OGBench benchmark when integrated with HIQL.

Rui Yang, Han Zhong, Jiawei Xu et al.

Offline reinforcement learning (RL) presents a promising approach for learning reinforced policies from offline datasets without the need for costly or unsafe interactions with the environment. However, datasets collected by humans in real-world environments are often noisy and may even be maliciously corrupted, which can significantly degrade the performance of offline RL. In this work, we first investigate the performance of current offline RL algorithms under comprehensive data corruption, including states, actions, rewards, and dynamics. Our extensive experiments reveal that implicit Q-learning (IQL) demonstrates remarkable resilience to data corruption among various offline RL algorithms. Furthermore, we conduct both empirical and theoretical analyses to understand IQL's robust performance, identifying its supervised policy learning scheme as the key factor. Despite its relative robustness, IQL still suffers from heavy-tail targets of Q functions under dynamics corruption. To tackle this challenge, we draw inspiration from robust statistics to employ the Huber loss to handle the heavy-tailedness and utilize quantile estimators to balance penalization for corrupted data and learning stability. By incorporating these simple yet effective modifications into IQL, we propose a more robust offline RL approach named Robust IQL (RIQL). Extensive experiments demonstrate that RIQL exhibits highly robust performance when subjected to diverse data corruption scenarios.

Brett Barkley, Amy Zhang, David Fridovich-Keil

One of the fundamental challenges associated with reinforcement learning (RL) is that collecting sufficient data can be both time-consuming and expensive. In this paper, we formalize a concept of time reversal symmetry in a Markov decision process (MDP), which builds upon the established structure of dynamically reversible Markov chains (DRMCs) and time-reversibility in classical physics. Specifically, we investigate the utility of this concept in reducing the sample complexity of reinforcement learning. We observe that utilizing the structure of time reversal in an MDP allows every environment transition experienced by an agent to be transformed into a feasible reverse-time transition, effectively doubling the number of experiences in the environment. To test the usefulness of this newly synthesized data, we develop a novel approach called time symmetric data augmentation (TSDA) and investigate its application in both proprioceptive and pixel-based state within the realm of off-policy, model-free RL. Empirical evaluations showcase how these synthetic transitions can enhance the sample efficiency of RL agents in time reversible scenarios without friction or contact. We also test this method in more realistic environments where these assumptions are not globally satisfied. We find that TSDA can significantly degrade sample efficiency and policy performance, but can also improve sample efficiency under the right conditions. Ultimately we conclude that time symmetry shows promise in enhancing the sample efficiency of reinforcement learning and provide guidance when the environment and reward structures are of an appropriate form for TSDA to be employed effectively.

Siddhant Agarwal, Ishan Durugkar, Peter Stone et al.

Goal-Conditioned Reinforcement Learning (RL) problems often have access to sparse rewards where the agent receives a reward signal only when it has achieved the goal, making policy optimization a difficult problem. Several works augment this sparse reward with a learned dense reward function, but this can lead to sub-optimal policies if the reward is misaligned. Moreover, recent works have demonstrated that effective shaping rewards for a particular problem can depend on the underlying learning algorithm. This paper introduces a novel way to encourage exploration called $f$-Policy Gradients, or $f$-PG. $f$-PG minimizes the f-divergence between the agent's state visitation distribution and the goal, which we show can lead to an optimal policy. We derive gradients for various f-divergences to optimize this objective. Our learning paradigm provides dense learning signals for exploration in sparse reward settings. We further introduce an entropy-regularized policy optimization objective, that we call $state$-MaxEnt RL (or $s$-MaxEnt RL) as a special case of our objective. We show that several metric-based shaping rewards like L2 can be used with $s$-MaxEnt RL, providing a common ground to study such metric-based shaping rewards with efficient exploration. We find that $f$-PG has better performance compared to standard policy gradient methods on a challenging gridworld as well as the Point Maze and FetchReach environments. More information on our website https://agarwalsiddhant10.github.io/projects/fpg.html.

Anuj Mahajan, Amy Zhang

Learning policies which are robust to changes in the environment are critical for real world deployment of Reinforcement Learning agents. They are also necessary for achieving good generalization across environment shifts. We focus on bisimulation metrics, which provide a powerful means for abstracting task relevant components of the observation and learning a succinct representation space for training the agent using reinforcement learning. In this work, we extend the bisimulation framework to also account for context dependent observation shifts. Specifically, we focus on the simulator based learning setting and use alternate observations to learn a representation space which is invariant to observation shifts using a novel bisimulation based objective. This allows us to deploy the agent to varying observation settings during test time and generalize to unseen scenarios. We further provide novel theoretical bounds for simulator fidelity and performance transfer guarantees for using a learnt policy to unseen shifts. Empirical analysis on the high-dimensional image based control domains demonstrates the efficacy of our method.

Mohamad Fares El Hajj Chehade, Amrit Singh Bedi, Amy Zhang et al.

Transfer learning in reinforcement learning (RL) has become a pivotal strategy for improving data efficiency in new, unseen tasks by utilizing knowledge from previously learned tasks. This approach is especially beneficial in real-world deployment scenarios where computational resources are constrained and agents must adapt rapidly to novel environments. However, current state-of-the-art methods often fall short in ensuring safety during the transfer process, particularly when unforeseen risks emerge in the deployment phase. In this work, we address these limitations by introducing a novel Caution-Aware Transfer Learning (CAT) framework. Unlike traditional approaches that limit risk considerations to mean-variance, we define "caution" as a more generalized and comprehensive notion of risk. Our core innovation lies in optimizing a weighted sum of reward return and caution-based on state-action occupancy measures-during the transfer process, allowing for a rich representation of diverse risk factors. To the best of our knowledge, this is the first work to explore the optimization of such a generalized risk notion within the context of transfer RL. Our contributions are threefold: (1) We propose a Caution-Aware Transfer (CAT) framework that evaluates source policies within the test environment and constructs a new policy that balances reward maximization and caution. (2) We derive theoretical sub-optimality bounds for our method, providing rigorous guarantees of its efficacy. (3) We empirically validate CAT, demonstrating that it consistently outperforms existing methods by delivering safer policies under varying risk conditions in the test tasks.

Zizhao Wang, Chang Shi, Jiaheng Hu et al.

Learning latent actions from action-free video has emerged as a powerful paradigm for scaling up controllable world model learning. Latent actions provide a natural interface for users to iteratively generate and manipulate videos. However, most existing approaches rely on monolithic inverse and forward dynamics models that learn a single latent action to control the entire scene, and therefore struggle in complex environments where multiple entities act simultaneously. This paper introduces Factored Latent Action Model (FLAM), a factored dynamics framework that decomposes the scene into independent factors, each inferring its own latent action and predicting its own next-step factor value. This factorized structure enables more accurate modeling of complex multi-entity dynamics and improves video generation quality in action-free video settings compared to monolithic models. Based on experiments on both simulation and real-world multi-entity datasets, we find that FLAM outperforms prior work in prediction accuracy and representation quality, and facilitates downstream policy learning, demonstrating the benefits of factorized latent action models.

Qinqing Zheng, Mikael Henaff, Amy Zhang et al.

Automatically synthesizing dense rewards from natural language descriptions is a promising paradigm in reinforcement learning (RL), with applications to sparse reward problems, open-ended exploration, and hierarchical skill design. Recent works have made promising steps by exploiting the prior knowledge of large language models (LLMs). However, these approaches suffer from important limitations: they are either not scalable to problems requiring billions of environment samples, due to requiring LLM annotations for each observation, or they require a diverse offline dataset, which may not exist or be impossible to collect. In this work, we address these limitations through a combination of algorithmic and systems-level contributions. We propose ONI, a distributed architecture that simultaneously learns an RL policy and an intrinsic reward function using LLM feedback. Our approach annotates the agent's collected experience via an asynchronous LLM server, which is then distilled into an intrinsic reward model. We explore a range of algorithmic choices for reward modeling with varying complexity, including hashing, classification, and ranking models. Our approach achieves state-of-the-art performance across a range of challenging tasks from the NetHack Learning Environment, while removing the need for large offline datasets required by prior work. We make our code available at https://github.com/facebookresearch/oni.

Shuozhe Li, Vaishnav Tadiparthi, Kwonjoon Lee et al.

Reinforcement learning from verifiable rewards (RLVR) produces strong reasoning models, yet they can fail catastrophically when the conditioning context is fallible (e.g., corrupted chain-of-thought, misleading partial solutions, or mild input perturbations), since standard RLVR optimizes final-answer correctness only under clean conditioning. We introduce GASP (Guided Adversarial Self-Play), a robustification method that explicitly trains detect-and-repair capabilities using only outcome verification. Without human labels or external teachers, GASP forms an adversarial self-play game within a single model: a polluter learns to induce failure via locally coherent corruptions, while an agent learns to diagnose and recover under the same corrupted conditioning. To address the scarcity of successful recoveries early in training, we propose in-distribution repair guidance, an imitation term on self-generated repairs that increases recovery probability while preserving previously acquired capabilities. Across four open-weight models (1.5B--8B), GASP transforms strong-but-brittle reasoners into robust ones that withstand misleading and perturbed context while often improving clean accuracy. Further analysis shows that adversarial corruptions induce an effective curriculum, and in-distribution guidance enables rapid recovery learning with minimal representational drift.

Alexander Levine, Peter Stone, Amy Zhang

In real-world control settings, the observation space is often unnecessarily high-dimensional and subject to time-correlated noise. However, the controllable dynamics of the system are often far simpler than the dynamics of the raw observations. It is therefore desirable to learn an encoder to map the observation space to a simpler space of control-relevant variables. In this work, we consider the Ex-BMDP model, first proposed by Efroni et al. (2022), which formalizes control problems where observations can be factorized into an action-dependent latent state which evolves deterministically, and action-independent time-correlated noise. Lamb et al. (2022) proposes the "AC-State" method for learning an encoder to extract a complete action-dependent latent state representation from the observations in such problems. AC-State is a multistep-inverse method, in that it uses the encoding of the the first and last state in a path to predict the first action in the path. However, we identify cases where AC-State will fail to learn a correct latent representation of the agent-controllable factor of the state. We therefore propose a new algorithm, ACDF, which combines multistep-inverse prediction with a latent forward model. ACDF is guaranteed to correctly infer an action-dependent latent state encoder for a large class of Ex-BMDP models. We demonstrate the effectiveness of ACDF on tabular Ex-BMDPs through numerical simulations; as well as high-dimensional environments using neural-network-based encoders. Code is available at https://github.com/midi-lab/acdf.

Carl Qi, Xiaojie Wang, Silong Yong et al.

Reasoning about failures is crucial for building reliable and trustworthy robotic systems. Prior approaches either treat failure reasoning as a closed-set classification problem or assume access to ample human annotations. Failures in the real world are typically subtle, combinatorial, and difficult to enumerate, whereas rich reasoning labels are expensive to acquire. We address this problem by introducing ARMOR: Adaptive Round-based Multi-task mOdel for Robotic failure detection and reasoning. We formulate detection and reasoning as a multi-task self-refinement process, where the model iteratively predicts detection outcomes and natural language reasoning conditioned on past outputs. During training, ARMOR learns from heterogeneous supervision - large-scale sparse binary labels and small-scale rich reasoning annotations - optimized via a combination of offline and online imitation learning. At inference time, ARMOR generates multiple refinement trajectories and selects the most confident prediction via a self-certainty metric. Experiments across diverse environments show that ARMOR achieves state-of-the-art performance by improving over the previous approaches by up to 30% on failure detection rate and up to 100% in reasoning measured through LLM fuzzy match score, demonstrating robustness to heterogeneous supervision and open-ended reasoning beyond predefined failure modes. We provide dditional visualizations on our website: https://sites.google.com/utexas.edu/armor

Max Rudolph, Nathan Lichtle, Sobhan Mohammadpour et al.

In the past decade, motivated by the putative failure of naive self-play deep reinforcement learning (DRL) in adversarial imperfect-information games, researchers have developed numerous DRL algorithms based on fictitious play (FP), double oracle (DO), and counterfactual regret minimization (CFR). In light of recent results of the magnetic mirror descent algorithm, we hypothesize that simpler generic policy gradient methods like PPO are competitive with or superior to these FP-, DO-, and CFR-based DRL approaches. To facilitate the resolution of this hypothesis, we implement and release the first broadly accessible exact exploitability computations for four large games. Using these games, we conduct the largest-ever exploitability comparison of DRL algorithms for imperfect-information games. Over 5600 training runs, we find that FP-, DO-, and CFR-based approaches fail to outperform generic policy gradient methods. Code is available at https://github.com/nathanlct/IIG-RL-Benchmark and https://github.com/gabrfarina/exp-a-spiel .

Haoran Xu, Kaiwen Hu, Somayeh Sojoudi et al.

We study behavior-regularized reinforcement learning (RL), where regularization toward a reference distribution (the dataset in offline RL or the base model in LLM RL finetuning) is essential to prevent value over-optimization caused by erroneous out-of-distribution extrapolation. Existing methods either rely on reparameterized policy gradient, which are difficult to scale to large generative models, or on reject sampling, which can be overly conservative when attempting to move beyond the behavior support. In this paper, we propose Value Gradient Flow (VGF), a scalable new paradigm for behavior-regularized RL. VGF casts behavior-regularized RL as an optimal transport problem that maps the reference distribution to the value-induced optimal policy distribution. We solve this transport problem via discrete gradient flow, where value gradients guide particles initialized from the reference distribution. Our analysis shows that VGF imposes regularization implicitly by controlling the transport budget. VGF eliminates explicit policy parameterization while remaining expressive and flexible, this enables adaptive test-time scaling by adjusting the transport budget. Extensive experiments demonstrate that VGF significantly outperforms prior methods, achieving state-of-the-art results on offline RL benchmarks (D4RL, OGBench) and LLM RL tasks. Code and runs can be found at https://ryanxhr.github.io/vgf.

Shuozhe Li, Jincheng Cao, Bodun Hu et al.

Reinforcement finetuning (RFT) has emerged as a powerful paradigm for unlocking reasoning capabilities in large language models. However, we identify a critical trade-off: while unconstrained RFT achieves strong reasoning performance, it severely compromises model trustworthiness by amplifying hallucination and worsening calibration; conversely, RKL-constrained RFT preserves trustworthiness but limits reasoning gains due to its unbounded penalty on exploratory deviations. To resolve this tension, we introduce CARE-RFT (Confidence-Anchored Regularized Reinforcement Finetuning), a novel method that replaces standard reverse KL regularization with a skew reverse KL divergence. CARE-RFT provides a confidence-sensitive penalty: it is bounded for confident, consistently rewarded explorations to enable reasoning, while unbounded elsewhere to preserve calibration. Extensive experiments across multiple model scales and RFT algorithms show that CARE-RFT achieves a superior balance, matching the reasoning performance of unconstrained RFT while recovering the trustworthiness and calibration of the base model. Our work establishes that careful, confidence-aware regularization is key to building both capable and trustworthy reasoning models.

Philippe Hansen-Estruch, Sriram Vishwanath, Amy Zhang et al.

At the core of both successful generative and self-supervised representation learning models there is a reconstruction objective that incorporates some form of image corruption. Diffusion models implement this approach through a scheduled Gaussian corruption process, while masked auto-encoder models do so by masking patches of the image. Despite their different approaches, the underlying similarity in their methodologies suggests a promising avenue for an auto-encoder capable of both de-noising tasks. We propose a unified self-supervised objective, dubbed Unified Masked Diffusion (UMD), that combines patch-based and noise-based corruption techniques within a single auto-encoding framework. Specifically, UMD modifies the diffusion transformer (DiT) training process by introducing an additional noise-free, high masking representation step in the diffusion noising schedule, and utilizes a mixed masked and noised image for subsequent timesteps. By integrating features useful for diffusion modeling and for predicting masked patch tokens, UMD achieves strong performance in downstream generative and representation learning tasks, including linear probing and class-conditional generation. This is achieved without the need for heavy data augmentations, multiple views, or additional encoders. Furthermore, UMD improves over the computational efficiency of prior diffusion based methods in total training time. We release our code at https://github.com/philippe-eecs/small-vision.

Luis Pineda, Brandon Amos, Amy Zhang et al.

Model-based reinforcement learning is a compelling framework for data-efficient learning of agents that interact with the world. This family of algorithms has many subcomponents that need to be carefully selected and tuned. As a result the entry-bar for researchers to approach the field and to deploy it in real-world tasks can be daunting. In this paper, we present MBRL-Lib -- a machine learning library for model-based reinforcement learning in continuous state-action spaces based on PyTorch. MBRL-Lib is designed as a platform for both researchers, to easily develop, debug and compare new algorithms, and non-expert user, to lower the entry-bar of deploying state-of-the-art algorithms. MBRL-Lib is open-source at https://github.com/facebookresearch/mbrl-lib.

Zihan Ding, Amy Zhang, Yuandong Tian et al.

We introduce Diffusion World Model (DWM), a conditional diffusion model capable of predicting multistep future states and rewards concurrently. As opposed to traditional one-step dynamics models, DWM offers long-horizon predictions in a single forward pass, eliminating the need for recursive queries. We integrate DWM into model-based value estimation, where the short-term return is simulated by future trajectories sampled from DWM. In the context of offline reinforcement learning, DWM can be viewed as a conservative value regularization through generative modeling. Alternatively, it can be seen as a data source that enables offline Q-learning with synthetic data. Our experiments on the D4RL dataset confirm the robustness of DWM to long-horizon simulation. In terms of absolute performance, DWM significantly surpasses one-step dynamics models with a $44\%$ performance gain, and is comparable to or slightly surpassing their model-free counterparts.

Scott Fujimoto, Pierluca D'Oro, Amy Zhang et al.

Reinforcement learning (RL) promises a framework for near-universal problem-solving. In practice however, RL algorithms are often tailored to specific benchmarks, relying on carefully tuned hyperparameters and algorithmic choices. Recently, powerful model-based RL methods have shown impressive general results across benchmarks but come at the cost of increased complexity and slow run times, limiting their broader applicability. In this paper, we attempt to find a unifying model-free deep RL algorithm that can address a diverse class of domains and problem settings. To achieve this, we leverage model-based representations that approximately linearize the value function, taking advantage of the denser task objectives used by model-based RL while avoiding the costs associated with planning or simulated trajectories. We evaluate our algorithm, MR.Q, on a variety of common RL benchmarks with a single set of hyperparameters and show a competitive performance against domain-specific and general baselines, providing a concrete step towards building general-purpose model-free deep RL algorithms.

Dongsu Lee, Daehee Lee, Amy Zhang

This work presents MAC-Flow, a simple yet expressive framework for multi-agent coordination. We argue that requirements of effective coordination are twofold: (i) a rich representation of the diverse joint behaviors present in offline data and (ii) the ability to act efficiently in real time. However, prior approaches often sacrifice one for the other, i.e., denoising diffusion-based solutions capture complex coordination but are computationally slow, while Gaussian policy-based solutions are fast but brittle in handling multi-agent interaction. MAC-Flow addresses this trade-off by first learning a flow-based representation of joint behaviors, and then distilling it into decentralized one-step policies that preserve coordination while enabling fast execution. Across four different benchmarks, including $12$ environments and $34$ datasets, MAC-Flow alleviates the trade-off between performance and computational cost, specifically achieving about $\boldsymbol{\times14.5}$ faster inference compared to diffusion-based MARL methods, while maintaining good performance. At the same time, its inference speed is similar to that of prior Gaussian policy-based offline multi-agent reinforcement learning (MARL) methods.

Jake Grigsby, Justin Sasek, Samyak Parajuli et al.

Language models trained on diverse datasets unlock generalization by in-context learning. Reinforcement Learning (RL) policies can achieve a similar effect by meta-learning within the memory of a sequence model. However, meta-RL research primarily focuses on adapting to minor variations of a single task. It is difficult to scale towards more general behavior without confronting challenges in multi-task optimization, and few solutions are compatible with meta-RL's goal of learning from large training sets of unlabeled tasks. To address this challenge, we revisit the idea that multi-task RL is bottlenecked by imbalanced training losses created by uneven return scales across different tasks. We build upon recent advancements in Transformer-based (in-context) meta-RL and evaluate a simple yet scalable solution where both an agent's actor and critic objectives are converted to classification terms that decouple optimization from the current scale of returns. Large-scale comparisons in Meta-World ML45, Multi-Game Procgen, Multi-Task POPGym, Multi-Game Atari, and BabyAI find that this design unlocks significant progress in online multi-task adaptation and memory problems without explicit task labels.

Tyler Ingebrand, Amy Zhang, Ufuk Topcu

Although reinforcement learning (RL) can solve many challenging sequential decision making problems, achieving zero-shot transfer across related tasks remains a challenge. The difficulty lies in finding a good representation for the current task so that the agent understands how it relates to previously seen tasks. To achieve zero-shot transfer, we introduce the function encoder, a representation learning algorithm which represents a function as a weighted combination of learned, non-linear basis functions. By using a function encoder to represent the reward function or the transition function, the agent has information on how the current task relates to previously seen tasks via a coherent vector representation. Thus, the agent is able to achieve transfer between related tasks at run time with no additional training. We demonstrate state-of-the-art data efficiency, asymptotic performance, and training stability in three RL fields by augmenting basic RL algorithms with a function encoder task representation.

Beth Goldberg, Diana Acosta-Navas, Michiel Bakker et al.

Two substantial technological advances have reshaped the public square in recent decades: first with the advent of the internet and second with the recent introduction of large language models (LLMs). LLMs offer opportunities for a paradigm shift towards more decentralized, participatory online spaces that can be used to facilitate deliberative dialogues at scale, but also create risks of exacerbating societal schisms. Here, we explore four applications of LLMs to improve digital public squares: collective dialogue systems, bridging systems, community moderation, and proof-of-humanity systems. Building on the input from over 70 civil society experts and technologists, we argue that LLMs both afford promising opportunities to shift the paradigm for conversations at scale and pose distinct risks for digital public squares. We lay out an agenda for future research and investments in AI that will strengthen digital public squares and safeguard against potential misuses of AI.

Martin Klissarov, Mikael Henaff, Roberta Raileanu et al.

Describing skills in natural language has the potential to provide an accessible way to inject human knowledge about decision-making into an AI system. We present MaestroMotif, a method for AI-assisted skill design, which yields high-performing and adaptable agents. MaestroMotif leverages the capabilities of Large Language Models (LLMs) to effectively create and reuse skills. It first uses an LLM's feedback to automatically design rewards corresponding to each skill, starting from their natural language description. Then, it employs an LLM's code generation abilities, together with reinforcement learning, for training the skills and combining them to implement complex behaviors specified in language. We evaluate MaestroMotif using a suite of complex tasks in the NetHack Learning Environment (NLE), demonstrating that it surpasses existing approaches in both performance and usability.

Zizhao Wang, Jiaheng Hu, Caleb Chuck et al.

Unsupervised skill discovery carries the promise that an intelligent agent can learn reusable skills through autonomous, reward-free environment interaction. Existing unsupervised skill discovery methods learn skills by encouraging distinguishable behaviors that cover diverse states. However, in complex environments with many state factors (e.g., household environments with many objects), learning skills that cover all possible states is impossible, and naively encouraging state diversity often leads to simple skills that are not ideal for solving downstream tasks. This work introduces Skill Discovery from Local Dependencies (Skild), which leverages state factorization as a natural inductive bias to guide the skill learning process. The key intuition guiding Skild is that skills that induce <b>diverse interactions</b> between state factors are often more valuable for solving downstream tasks. To this end, Skild develops a novel skill learning objective that explicitly encourages the mastering of skills that effectively induce different interactions within an environment. We evaluate Skild in several domains with challenging, long-horizon sparse reward tasks including a realistic simulated household robot domain, where Skild successfully learns skills with clear semantic meaning and shows superior performance compared to existing unsupervised reinforcement learning methods that only maximize state coverage.

Harshit Sikchi, Andrea Tirinzoni, Ahmed Touati et al.

Unsupervised zero-shot reinforcement learning (RL) has emerged as a powerful paradigm for pretraining behavioral foundation models (BFMs), enabling agents to solve a wide range of downstream tasks specified via reward functions in a zero-shot fashion, i.e., without additional test-time learning or planning. This is achieved by learning self-supervised task embeddings alongside corresponding near-optimal behaviors and incorporating an inference procedure to directly retrieve the latent task embedding and associated policy for any given reward function. Despite promising results, zero-shot policies are often suboptimal due to errors induced by the unsupervised training process, the embedding, and the inference procedure. In this paper, we focus on devising fast adaptation strategies to improve the zero-shot performance of BFMs in a few steps of online interaction with the environment while avoiding any performance drop during the adaptation process. Notably, we demonstrate that existing BFMs learn a set of skills containing more performant policies than those identified by their inference procedure, making them well-suited for fast adaptation. Motivated by this observation, we propose both actor-critic and actor-only fast adaptation strategies that search in the low-dimensional task-embedding space of the pre-trained BFM to rapidly improve the performance of its zero-shot policies on any downstream task. Notably, our approach mitigates the initial "unlearning" phase commonly observed when fine-tuning pre-trained RL models. We evaluate our fast adaptation strategies on top of four state-of-the-art zero-shot RL methods in multiple navigation and locomotion domains. Our results show that they achieve 10-40% improvement over their zero-shot performance in a few tens of episodes, outperforming existing baselines.

Ruiyang Zhou, Shuozhe Li, Amy Zhang et al.

Recent advances in large language models have been driven by reinforcement learning (RL)-style post-training, which improves reasoning by optimizing model outputs based on reward or preference signals. GRPO-style approaches implement this by using self-generated samples labeled by an outcome-based verifier. However, these methods depend heavily on the model's initial ability to produce positive samples. They primarily refine what the model already knows (distribution sharpening) rather than enabling the model to solve problems where it initially fails. This limitation is especially problematic in early-stage RL training and on challenging reasoning tasks, where positive samples are unlikely to be generated. To unlock reasoning ability in such settings, the model must explore new reasoning trajectories beyond its current output distribution. Such exploration requires access to sufficiently good positive samples to guide the learning. While expert demonstrations seem like a natural solution, we find that they are often ineffective in RL post-training. Instead, we identify two key properties of effective positive samples: they should (1) be likely under the current policy, and (2) increase the model's likelihood of predicting the correct answer. Based on these insights, we propose $\textbf{Self-Explanation Policy Optimization (ExPO)}$-a simple and modular framework that generates such samples by conditioning on the ground-truth answer. ExPO enables efficient exploration and guides the model to produce reasoning trajectories more aligned with its policy than expert-written CoTs, while ensuring higher quality than its own (incorrect) samples. Experiments show that ExPO improves both learning efficiency and final performance on reasoning benchmarks, surpassing expert-demonstration-based methods in challenging settings such as MATH level-5, where the model initially struggles the most.

Arthur Zhang, Harshit Sikchi, Amy Zhang et al.

We introduce CREStE, a scalable learning-based mapless navigation framework to address the open-world generalization and robustness challenges of outdoor urban navigation. Key to achieving this is learning perceptual representations that generalize to open-set factors (e.g. novel semantic classes, terrains, dynamic entities) and inferring expert-aligned navigation costs from limited demonstrations. CREStE addresses both these issues, introducing 1) a visual foundation model (VFM) distillation objective for learning open-set structured bird's-eye-view perceptual representations, and 2) counterfactual inverse reinforcement learning (IRL), a novel active learning formulation that uses counterfactual trajectory demonstrations to reason about the most important cues when inferring navigation costs. We evaluate CREStE on the task of kilometer-scale mapless navigation in a variety of city, offroad, and residential environments and find that it outperforms all state-of-the-art approaches with 70% fewer human interventions, including a 2-kilometer mission in an unseen environment with just 1 intervention; showcasing its robustness and effectiveness for long-horizon mapless navigation. Videos and additional materials can be found on the project page: https://amrl.cs.utexas.edu/creste

Carl Qi, Dan Haramati, Tal Daniel et al.

Object manipulation is a common component of everyday tasks, but learning to manipulate objects from high-dimensional observations presents significant challenges. These challenges are heightened in multi-object environments due to the combinatorial complexity of the state space as well as of the desired behaviors. While recent approaches have utilized large-scale offline data to train models from pixel observations, achieving performance gains through scaling, these methods struggle with compositional generalization in unseen object configurations with constrained network and dataset sizes. To address these issues, we propose a novel behavioral cloning (BC) approach that leverages object-centric representations and an entity-centric Transformer with diffusion-based optimization, enabling efficient learning from offline image data. Our method first decomposes observations into an object-centric representation, which is then processed by our entity-centric Transformer that computes attention at the object level, simultaneously predicting object dynamics and the agent's actions. Combined with the ability of diffusion models to capture multi-modal behavior distributions, this results in substantial performance improvements in multi-object tasks and, more importantly, enables compositional generalization. We present BC agents capable of zero-shot generalization to tasks with novel compositions of objects and goals, including larger numbers of objects than seen during training. We provide video rollouts on our webpage: https://sites.google.com/view/ec-diffuser.

Harshit Sikchi, Siddhant Agarwal, Pranaya Jajoo et al.

The reward hypothesis states that all goals and purposes can be understood as the maximization of a received scalar reward signal. However, in practice, defining such a reward signal is notoriously difficult, as humans are often unable to predict the optimal behavior corresponding to a reward function. Natural language offers an intuitive alternative for instructing reinforcement learning (RL) agents, yet previous language-conditioned approaches either require costly supervision or test-time training given a language instruction. In this work, we present a new approach that uses a pretrained RL agent trained using only unlabeled, offline interactions--without task-specific supervision or labeled trajectories--to get zero-shot test-time policy inference from arbitrary natural language instructions. We introduce a framework comprising three steps: imagine, project, and imitate. First, the agent imagines a sequence of observations corresponding to the provided language description using video generative models. Next, these imagined observations are projected into the target environment domain. Finally, an agent pretrained in the target environment with unsupervised RL instantly imitates the projected observation sequence through a closed-form solution. To the best of our knowledge, our method, RLZero, is the first approach to show direct language-to-behavior generation abilities on a variety of tasks and environments without any in-domain supervision. We further show that components of RLZero can be used to generate policies zero-shot from cross-embodied videos, such as those available on YouTube, even for complex embodiments like humanoids.