Scott Niekum

W. Bradley Knox, Stephane Hatgis-Kessell, Serena Booth et al.

The utility of reinforcement learning is limited by the alignment of reward functions with the interests of human stakeholders. One promising method for alignment is to learn the reward function from human-generated preferences between pairs of trajectory segments, a type of reinforcement learning from human feedback (RLHF). These human preferences are typically assumed to be informed solely by partial return, the sum of rewards along each segment. We find this assumption to be flawed and propose modeling human preferences instead as informed by each segment's regret, a measure of a segment's deviation from optimal decision-making. Given infinitely many preferences generated according to regret, we prove that we can identify a reward function equivalent to the reward function that generated those preferences, and we prove that the previous partial return model lacks this identifiability property in multiple contexts. We empirically show that our proposed regret preference model outperforms the partial return preference model with finite training data in otherwise the same setting. Additionally, we find that our proposed regret preference model better predicts real human preferences and also learns reward functions from these preferences that lead to policies that are better human-aligned. Overall, this work establishes that the choice of preference model is impactful, and our proposed regret preference model provides an improvement upon a core assumption of recent research. We have open sourced our experimental code, the human preferences dataset we gathered, and our training and preference elicitation interfaces for gathering a such a dataset.

Joey Hejna, Rafael Rafailov, Harshit Sikchi et al.

Reinforcement Learning from Human Feedback (RLHF) has emerged as a popular paradigm for aligning models with human intent. Typically RLHF algorithms operate in two phases: first, use human preferences to learn a reward function and second, align the model by optimizing the learned reward via reinforcement learning (RL). This paradigm assumes that human preferences are distributed according to reward, but recent work suggests that they instead follow the regret under the user's optimal policy. Thus, learning a reward function from feedback is not only based on a flawed assumption of human preference, but also leads to unwieldy optimization challenges that stem from policy gradients or bootstrapping in the RL phase. Because of these optimization challenges, contemporary RLHF methods restrict themselves to contextual bandit settings (e.g., as in large language models) or limit observation dimensionality (e.g., state-based robotics). We overcome these limitations by introducing a new family of algorithms for optimizing behavior from human feedback using the regret-based model of human preferences. Using the principle of maximum entropy, we derive Contrastive Preference Learning (CPL), an algorithm for learning optimal policies from preferences without learning reward functions, circumventing the need for RL. CPL is fully off-policy, uses only a simple contrastive objective, and can be applied to arbitrary MDPs. This enables CPL to elegantly scale to high-dimensional and sequential RLHF problems while being simpler than prior methods.

Yuchen Cui, Scott Niekum, Abhinav Gupta et al.

Task specification is at the core of programming autonomous robots. A low-effort modality for task specification is critical for engagement of non-expert end-users and ultimate adoption of personalized robot agents. A widely studied approach to task specification is through goals, using either compact state vectors or goal images from the same robot scene. The former is hard to interpret for non-experts and necessitates detailed state estimation and scene understanding. The latter requires the generation of desired goal image, which often requires a human to complete the task, defeating the purpose of having autonomous robots. In this work, we explore alternate and more general forms of goal specification that are expected to be easier for humans to specify and use such as images obtained from the internet, hand sketches that provide a visual description of the desired task, or simple language descriptions. As a preliminary step towards this, we investigate the capabilities of large scale pre-trained models (foundation models) for zero-shot goal specification, and find promising results in a collection of simulated robot manipulation tasks and real-world datasets.

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.

Prasoon Goyal, Raymond J. Mooney, Scott Niekum

We introduce a novel setting, wherein an agent needs to learn a task from a demonstration of a related task with the difference between the tasks communicated in natural language. The proposed setting allows reusing demonstrations from other tasks, by providing low effort language descriptions, and can also be used to provide feedback to correct agent errors, which are both important desiderata for building intelligent agents that assist humans in daily tasks. To enable progress in this proposed setting, we create two benchmarks -- Room Rearrangement and Room Navigation -- that cover a diverse set of task adaptations. Further, we propose a framework that uses a transformer-based model to reason about the entities in the tasks and their relationships, to learn a policy for the target task

Wonjoon Goo, Scott Niekum

We introduce an offline reinforcement learning (RL) algorithm that explicitly clones a behavior policy to constrain value learning. In offline RL, it is often important to prevent a policy from selecting unobserved actions, since the consequence of these actions cannot be presumed without additional information about the environment. One straightforward way to implement such a constraint is to explicitly model a given data distribution via behavior cloning and directly force a policy not to select uncertain actions. However, many offline RL methods instantiate the constraint indirectly -- for example, pessimistic value estimation -- due to a concern about errors when modeling a potentially complex behavior policy. In this work, we argue that it is not only viable but beneficial to explicitly model the behavior policy for offline RL because the constraint can be realized in a stable way with the trained model. We first suggest a theoretical framework that allows us to incorporate behavior-cloned models into value-based offline RL methods, enjoying the strength of both explicit behavior cloning and value learning. Then, we propose a practical method utilizing a score-based generative model for behavior cloning. With the proposed method, we show state-of-the-art performance on several datasets within the D4RL and Robomimic benchmarks and achieve competitive performance across all datasets tested.

W. Bradley Knox, Stephane Hatgis-Kessell, Sigurdur Orn Adalgeirsson et al.

We consider algorithms for learning reward functions from human preferences over pairs of trajectory segments, as used in reinforcement learning from human feedback (RLHF). Most recent work assumes that human preferences are generated based only upon the reward accrued within those segments, or their partial return. Recent work casts doubt on the validity of this assumption, proposing an alternative preference model based upon regret. We investigate the consequences of assuming preferences are based upon partial return when they actually arise from regret. We argue that the learned function is an approximation of the optimal advantage function, $\hat{A^*_r}$, not a reward function. We find that if a specific pitfall is addressed, this incorrect assumption is not particularly harmful, resulting in a highly shaped reward function. Nonetheless, this incorrect usage of $\hat{A^*_r}$ is less desirable than the appropriate and simpler approach of greedy maximization of $\hat{A^*_r}$. From the perspective of the regret preference model, we also provide a clearer interpretation of fine tuning contemporary large language models with RLHF. This paper overall provides insight regarding why learning under the partial return preference model tends to work so well in practice, despite it conforming poorly to how humans give preferences.

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.

Andrew Levy, Sreehari Rammohan, Alessandro Allievi et al.

General purpose agents will require large repertoires of skills. Empowerment -- the maximum mutual information between skills and states -- provides a pathway for learning large collections of distinct skills, but mutual information is difficult to optimize. We introduce a new framework, Hierarchical Empowerment, that makes computing empowerment more tractable by integrating concepts from Goal-Conditioned Hierarchical Reinforcement Learning. Our framework makes two specific contributions. First, we introduce a new variational lower bound on mutual information that can be used to compute empowerment over short horizons. Second, we introduce a hierarchical architecture for computing empowerment over exponentially longer time scales. We verify the contributions of the framework in a series of simulated robotics tasks. In a popular ant navigation domain, our four level agents are able to learn skills that cover a surface area over two orders of magnitude larger than prior work.

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.

Ameya Agaskar, Sriram Siva, William Pickering et al.

We introduce DeepFleet, a suite of foundation models designed to support coordination and planning for large-scale mobile robot fleets. These models are trained on fleet movement data, including robot positions, goals, and interactions, from hundreds of thousands of robots in Amazon warehouses worldwide. DeepFleet consists of four architectures that each embody a distinct inductive bias and collectively explore key points in the design space for multi-agent foundation models: the robot-centric (RC) model is an autoregressive decision transformer operating on neighborhoods of individual robots; the robot-floor (RF) model uses a transformer with cross-attention between robots and the warehouse floor; the image-floor (IF) model applies convolutional encoding to a multi-channel image representation of the full fleet; and the graph-floor (GF) model combines temporal attention with graph neural networks for spatial relationships. In this paper, we describe these models and present our evaluation of the impact of these design choices on prediction task performance. We find that the robot-centric and graph-floor models, which both use asynchronous robot state updates and incorporate the localized structure of robot interactions, show the most promise. We also present experiments that show that these two models can make effective use of larger warehouses operation datasets as the models are scaled up.

Caleb Chuck, Kevin Black, Aditya Arjun et al.

Reinforcement Learning (RL) has demonstrated promising results in learning policies for complex tasks, but it often suffers from low sample efficiency and limited transferability. Hierarchical RL (HRL) methods aim to address the difficulty of learning long-horizon tasks by decomposing policies into skills, abstracting states, and reusing skills in new tasks. However, many HRL methods require some initial task success to discover useful skills, which paradoxically may be very unlikely without access to useful skills. On the other hand, reward-free HRL methods often need to learn far too many skills to achieve proper coverage in high-dimensional domains. In contrast, we introduce the Chain of Interaction Skills (COInS) algorithm, which focuses on controllability in factored domains to identify a small number of task-agnostic skills that still permit a high degree of control. COInS uses learned detectors to identify interactions between state factors and then trains a chain of skills to control each of these factors successively. We evaluate COInS on a robotic pushing task with obstacles -- a challenging domain where other RL and HRL methods fall short. We also demonstrate the transferability of skills learned by COInS, using variants of Breakout, a common RL benchmark, and show 2-3x improvement in both sample efficiency and final performance compared to standard RL baselines.

Yaswanth Chittepu, Ativ Joshi, Rajarshi Bhattacharjee et al.

Safe Reinforcement Learning from Human Feedback (RLHF) typically enforces safety through expected cost constraints, but the expectation captures only a single statistic of the cost distribution and fails to account for distributional uncertainty, particularly under heavy tails or rare catastrophic events. This limitation is problematic when robustness and risk sensitivity are critical. Stochastic dominance offers a principled alternative by comparing entire cost distributions rather than just their averages, enabling direct control over tail risks and potential out-of-distribution failures that expectation-based constraints may overlook. In this work, we propose Risk-sensitive Alignment via Dominance (RAD), a novel alignment framework that replaces scalar expected cost constraints with First-Order Stochastic Dominance (FSD) constraints. We operationalize this constraint by comparing the target policy's cost distribution to that of a reference policy within an Optimal Transport (OT) framework, using entropic regularization and Sinkhorn iterations to obtain a differentiable and computationally efficient objective for stable end-to-end optimization. Furthermore, we introduce quantile-weighted FSD constraints and show that weighted FSD universally controls a broad class of Spectral Risk Measures (SRMs), so that improvements under weighted dominance imply guaranteed improvements in the corresponding spectral risk. This provides a principled mechanism for tuning a model's risk profile via the quantile weighting function. Empirical results demonstrate that RAD improves harmlessness over baselines while remaining competitive in helpfulness, and exhibits greater robustness on out-of-distribution harmlessness evaluations.

Will Schwarzer, Scott Niekum

Estimating how often an ML model will fail at deployment scale is central to pre-deployment safety assessment, but a feasible evaluation set is rarely large enough to observe the failures that matter. Jones et al. (2025) address this by extrapolating from the largest k failure scores in an evaluation set to predict deployment-scale failure rates. We give a finite-k decomposition of this estimator's forecast error and show that it has a built-in bias toward over-prediction in the typical case, which is the safety-favorable direction. This bias is offset when the evaluation set misses a rare high-failure mode that the deployment set contains, leaving the forecast to under-predict at deployment scale. We propose a fine-tuning objective, the forecastability loss, that addresses this failure mode. In two proof-of-concept experiments, a language-model password game and an RL gridworld, fine-tuning substantially reduces held-out forecast error while preserving primary-task capability and achieving safety similar to that of supervised baselines.

Daniel S. Brown, Scott Niekum, Marek Petrik

One of the main challenges in imitation learning is determining what action an agent should take when outside the state distribution of the demonstrations. Inverse reinforcement learning (IRL) can enable generalization to new states by learning a parameterized reward function, but these approaches still face uncertainty over the true reward function and corresponding optimal policy. Existing safe imitation learning approaches based on IRL deal with this uncertainty using a maxmin framework that optimizes a policy under the assumption of an adversarial reward function, whereas risk-neutral IRL approaches either optimize a policy for the mean or MAP reward function. While completely ignoring risk can lead to overly aggressive and unsafe policies, optimizing in a fully adversarial sense is also problematic as it can lead to overly conservative policies that perform poorly in practice. To provide a bridge between these two extremes, we propose Bayesian Robust Optimization for Imitation Learning (BROIL). BROIL leverages Bayesian reward function inference and a user specific risk tolerance to efficiently optimize a robust policy that balances expected return and conditional value at risk. Our empirical results show that BROIL provides a natural way to interpolate between return-maximizing and risk-minimizing behaviors and outperforms existing risk-sensitive and risk-neutral inverse reinforcement learning algorithms. Code is available at https://github.com/dsbrown1331/broil.

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.

Prasann Singhal, Nathan Lambert, Scott Niekum et al.

Varied approaches for aligning language models have been proposed, including supervised fine-tuning, RLHF, and direct optimization methods such as DPO. Although DPO has rapidly gained popularity due to its straightforward training process and competitive results, there is an open question of whether there remain practical advantages of using a discriminator, like a reward model, to evaluate responses. We propose D2PO, discriminator-guided DPO, an approach for the online setting where preferences are being collected throughout learning. As we collect gold preferences, we use these not only to train our policy, but to train a discriminative response evaluation model to silver-label even more synthetic data for policy training. We explore this approach across a set of diverse tasks, including a realistic chat setting, we find that our approach leads to higher-quality outputs compared to DPO with the same data budget, and greater efficiency in terms of preference data requirements. Furthermore, we show conditions under which silver labeling is most helpful: it is most effective when training the policy with DPO, outperforming traditional PPO, and benefits from maintaining a separate discriminator from the policy model.

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.

Tuhina Tripathi, Manya Wadhwa, Greg Durrett et al.

Large Language Models (LLMs) are widely used as proxies for human labelers in both training (Reinforcement Learning from AI Feedback) and large-scale response evaluation (LLM-as-a-judge). Alignment and evaluation are critical components in the development of reliable LLMs, and the choice of feedback protocol plays a central role in both but remains understudied. In this work, we show that the choice of feedback protocol for evaluation (absolute scores versus relative preferences) can significantly affect evaluation reliability and induce systematic biases. In the context of LLM-as-a-judge evaluation, we show that pairwise protocols are more vulnerable to distracted evaluation. Generator models can exploit spurious attributes (or distractor features) favored by the LLM judge, resulting in inflated scores for lower-quality outputs. We find that absolute scoring is more robust to such manipulation, producing judgments that better reflect response quality and are less influenced by distractor features. Our results demonstrate that generator models can flip preferences by embedding distractor features, skewing LLM-as-a-judge comparisons and leading to inaccurate conclusions about model quality in benchmark evaluations. Pairwise preferences flip in about 35% of the cases, compared to only 9% for absolute scores. We offer recommendations for choosing feedback protocols based on dataset characteristics and evaluation objectives.

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.

Caleb Chuck, Sankaran Vaidyanathan, Stephen Giguere et al.

Reinforcement learning (RL) algorithms often struggle to learn policies that generalize to novel situations due to issues such as causal confusion, overfitting to irrelevant factors, and failure to isolate control of state factors. These issues stem from a common source: a failure to accurately identify and exploit state-specific causal relationships in the environment. While some prior works in RL aim to identify these relationships explicitly, they rely on informal domain-specific heuristics such as spatial and temporal proximity. Actual causality offers a principled and general framework for determining the causes of particular events. However, existing definitions of actual cause often attribute causality to a large number of events, even if many of them rarely influence the outcome. Prior work on actual causality proposes normality as a solution to this problem, but its existing implementations are challenging to scale to complex and continuous-valued RL environments. This paper introduces functional actual cause (FAC), a framework that uses context-specific independencies in the environment to restrict the set of actual causes. We additionally introduce Joint Optimization for Actual Cause Inference (JACI), an algorithm that learns from observational data to infer functional actual causes. We demonstrate empirically that FAC agrees with known results on a suite of examples from the actual causality literature, and JACI identifies actual causes with significantly higher accuracy than existing heuristic methods in a set of complex, continuous-valued environments.

Stephen Chung, Scott Niekum, David Krueger

As reinforcement learning agents become increasingly deployed in real-world scenarios, predicting future agent actions and events during deployment is important for facilitating better human-agent interaction and preventing catastrophic outcomes. This paper experimentally evaluates and compares the effectiveness of future action and event prediction for three types of RL agents: explicitly planning, implicitly planning, and non-planning. We employ two approaches: the inner state approach, which involves predicting based on the inner computations of the agents (e.g., plans or neuron activations), and a simulation-based approach, which involves unrolling the agent in a learned world model. Our results show that the plans of explicitly planning agents are significantly more informative for prediction than the neuron activations of the other types. Furthermore, using internal plans proves more robust to model quality compared to simulation-based approaches when predicting actions, while the results for event prediction are more mixed. These findings highlight the benefits of leveraging inner states and simulations to predict future agent actions and events, thereby improving interaction and safety in real-world deployments.

Yaswanth Chittepu, Blossom Metevier, Will Schwarzer et al.

Existing approaches to language model alignment often treat safety as a tradeoff against helpfulness, which can lead to unacceptable responses in sensitive domains. To ensure reliable performance in such settings, we propose High-Confidence Safe Reinforcement Learning from Human Feedback (HC-RLHF), a method that provides high-confidence safety guarantees while maximizing helpfulness. Similar to previous methods, HC-RLHF explicitly decouples human preferences into helpfulness and harmlessness (safety), which are learned by training a reward model and a cost model, respectively. It then employs a two-step process to find safe solutions. In the first step, it optimizes the reward function under an intentionally pessimistic version of the cost constraint. In the second step, the trained model undergoes a safety test to verify whether its performance stays within an upper-confidence bound of the actual cost constraint. We provide a theoretical analysis of HC-RLHF, including proof that it will not return an unsafe solution with a probability greater than a user-specified threshold. For our empirical analysis, we apply HC-RLHF to align three different language models (Qwen2-1.5B, Qwen2.5-3B, and LLaMa3.2-3B) with human preferences. Our results demonstrate that HC-RLHF produces safe models with high probability and can improve harmlessness and helpfulness compared to previous methods.

Caleb Chuck, Fan Feng, Carl Qi et al.

Hindsight relabeling is a powerful tool for overcoming sparsity in goal-conditioned reinforcement learning (GCRL), especially in certain domains such as navigation and locomotion. However, hindsight relabeling can struggle in object-centric domains. For example, suppose that the goal space consists of a robotic arm pushing a particular target block to a goal location. In this case, hindsight relabeling will give high rewards to any trajectory that does not interact with the block. However, these behaviors are only useful when the object is already at the goal -- an extremely rare case in practice. A dataset dominated by these kinds of trajectories can complicate learning and lead to failures. In object-centric domains, one key intuition is that meaningful trajectories are often characterized by object-object interactions such as pushing the block with the gripper. To leverage this intuition, we introduce Hindsight Relabeling using Interactions (HInt), which combines interactions with hindsight relabeling to improve the sample efficiency of downstream RL. However because interactions do not have a consensus statistical definition tractable for downstream GCRL, we propose a definition of interactions based on the concept of null counterfactual: a cause object is interacting with a target object if, in a world where the cause object did not exist, the target object would have different transition dynamics. We leverage this definition to infer interactions in Null Counterfactual Interaction Inference (NCII), which uses a "nulling'' operation with a learned model to infer interactions. NCII is able to achieve significantly improved interaction inference accuracy in both simple linear dynamics domains and dynamic robotic domains in Robosuite, Robot Air Hockey, and Franka Kitchen and HInt improves sample efficiency by up to 4x.

Yaswanth Chittepu, Prasann Singhal, Greg Durrett et al.

Margin-based optimization is fundamental to improving generalization and robustness in classification tasks. In the context of reward model learning from preferences within Reinforcement Learning from Human Feedback (RLHF), existing methods typically rely on no margins, fixed margins, or margins that are simplistic functions of preference ratings. However, such formulations often fail to account for the varying strengths of different preferences, for example some preferences are associated with larger margins between responses, or they rely on noisy margin information derived from ratings. We argue that modeling the strength of preferences can lead to better generalization and more faithful alignment. Furthermore, many existing methods that use adaptive margins assume access to accurate preference scores, which can be difficult for humans to provide reliably. We propose an approach that leverages preferences over preferences, that is annotations indicating which of two preferences reflects a stronger distinction. We use this ordinal signal to infer adaptive margins on a per-datapoint basis. We introduce an extension to Direct Preference Optimization (DPO), DPO-PoP, that incorporates adaptive margins from preference-over-preference supervision, enabling improved discriminative and generative performance. Empirically, our method outperforms vanilla DPO, DPO with fixed margins, and DPO with ground-truth margins on the UltraFeedback dataset. Additionally, we show that there is a tradeoff between discriminative and generative performance: improving test classification accuracy, particularly by correctly labeling weaker preferences at the expense of stronger ones, can lead to a decline in generative quality. To navigate this tradeoff, we propose two sampling strategies to gather preference-over-preference labels: one favoring discriminative performance and one favoring generative performance.

Haoran Xu, Shuozhe Li, Harshit Sikchi et al.

We introduce Iterative Dual Reinforcement Learning (IDRL), a new method that takes an optimal discriminator-weighted imitation view of solving RL. Our method is motivated by a simple experiment in which we find training a discriminator using the offline dataset plus an additional expert dataset and then performing discriminator-weighted behavior cloning gives strong results on various types of datasets. That optimal discriminator weight is quite similar to the learned visitation distribution ratio in Dual-RL, however, we find that current Dual-RL methods do not correctly estimate that ratio. In IDRL, we propose a correction method to iteratively approach the optimal visitation distribution ratio in the offline dataset given no addtional expert dataset. During each iteration, IDRL removes zero-weight suboptimal transitions using the learned ratio from the previous iteration and runs Dual-RL on the remaining subdataset. This can be seen as replacing the behavior visitation distribution with the optimized visitation distribution from the previous iteration, which theoretically gives a curriculum of improved visitation distribution ratios that are closer to the optimal discriminator weight. We verify the effectiveness of IDRL on various kinds of offline datasets, including D4RL datasets and more realistic corrupted demonstrations. IDRL beats strong Primal-RL and Dual-RL baselines in terms of both performance and stability, on all datasets.

Stephane Hatgis-Kessell, W. Bradley Knox, Serena Booth et al.

Designing a reinforcement learning from human feedback (RLHF) algorithm to approximate a human's unobservable reward function requires assuming, implicitly or explicitly, a model of human preferences. A preference model that poorly describes how humans generate preferences risks learning a poor approximation of the human's reward function. In this paper, we conduct three human studies to asses whether one can influence the expression of real human preferences to more closely conform to a desired preference model. Importantly, our approach does not seek to alter the human's unobserved reward function. Rather, we change how humans use this reward function to generate preferences, such that they better match whatever preference model is assumed by a particular RLHF algorithm. We introduce three interventions: showing humans the quantities that underlie a preference model, which is normally unobservable information derived from the reward function; training people to follow a specific preference model; and modifying the preference elicitation question. All intervention types show significant effects, providing practical tools to improve preference data quality and the resultant alignment of the learned reward functions. Overall we establish a novel research direction in model alignment: designing interfaces and training interventions to increase human conformance with the modeling assumptions of the algorithm that will learn from their input.

Max Rudolph, Caleb Chuck, Kevin Black et al.

Robust reinforcement learning agents using high-dimensional observations must be able to identify relevant state features amidst many exogeneous distractors. A representation that captures controllability identifies these state elements by determining what affects agent control. While methods such as inverse dynamics and mutual information capture controllability for a limited number of timesteps, capturing long-horizon elements remains a challenging problem. Myopic controllability can capture the moment right before an agent crashes into a wall, but not the control-relevance of the wall while the agent is still some distance away. To address this we introduce action-bisimulation encoding, a method inspired by the bisimulation invariance pseudometric, that extends single-step controllability with a recursive invariance constraint. By doing this, action-bisimulation learns a multi-step controllability metric that smoothly discounts distant state features that are relevant for control. We demonstrate that action-bisimulation pretraining on reward-free, uniformly random data improves sample efficiency in several environments, including a photorealistic 3D simulation domain, Habitat. Additionally, we provide theoretical analysis and qualitative results demonstrating the information captured by action-bisimulation.

Shripad Vilasrao Deshmukh, Will Schwarzer, Scott Niekum

Policy evaluation is often a prerequisite for deploying safety- and performance-critical systems. Existing evaluation approaches frequently suffer from high variance due to limited data and long-horizon tasks, or high bias due to unequal support or inaccurate environmental models. We posit that these challenges arise, in part, from the standard reinforcement learning (RL) paradigm of policy learning without explicit consideration of evaluation. As an alternative, we propose evaluation-aware reinforcement learning (EvA-RL), in which a policy is trained to maximize expected return while simultaneously minimizing expected evaluation error under a given value prediction scheme -- in other words, being "easy" to evaluate. We formalize a framework for EvA-RL and design an instantiation that enables accurate policy evaluation, conditioned on a small number of rollouts in an assessment environment that can be different than the deployment environment. However, our theoretical analysis and empirical results show that there is often a tradeoff between evaluation accuracy and policy performance when using a fixed value-prediction scheme within EvA-RL. To mitigate this tradeoff, we extend our approach to co-learn an assessment-conditioned state-value predictor alongside the policy. Empirical results across diverse discrete and continuous action domains demonstrate that EvA-RL can substantially reduce evaluation error while maintaining competitive returns. This work lays the foundation for a broad new class of RL methods that treat reliable evaluation as a first-class principle during training.

Sylee Dandekar, Shripad Deshmukh, Frank Chiu et al.

Human preferences in RLHF are typically modeled as a function of the human's reward function or corresponding optimal state-action values. In this work, we propose that human beliefs about the capabilities of the agent being trained also play a key role in preference generation. We examine two questions related to this hypothesis, one descriptive and one normative, respectively: Do human labelers' beliefs about agent capabilities affect the preferences that they provide? And what is the ideal set of beliefs about an agent -- and resulting preferences -- for humans to have? We propose a new preference model that incorporates human beliefs and provide a normative theory that bounds the error on the final learned policy based on the \textit{mismatch} between the human's beliefs and an idealized set of beliefs. We then confirm via a human study that beliefs about agent capabilities do, in fact, significantly affect preferences and can be influenced through simple interventions. Additionally, we empirically show through synthetic experiments that it is often suboptimal for human preference labelers to assume agent optimality. Collectively, these results theoretically and empirically demonstrate how reducing the mismatch between human beliefs and agent capabilities can lead to more performant RLHF and point toward new best practices for RLHF practitioners.

Will Schwarzer, Jordan Schneider, Philip S. Thomas et al.

Existing approaches to reward inference from behavior typically assume that humans provide demonstrations according to specific models of behavior. However, humans often indicate their goals through a wide range of behaviors, from actions that are suboptimal due to poor planning or execution to behaviors which are intended to communicate goals rather than achieve them. We propose that supervised learning offers a unified framework to infer reward functions from any class of behavior, and show that such an approach is asymptotically Bayes-optimal under mild assumptions. Experiments on simulated robotic manipulation tasks show that our method can efficiently infer rewards from a wide variety of arbitrarily suboptimal demonstrations.

Ryan Bahlous-Boldi, Li Ding, Lee Spector et al.

Ensuring AI models align with human values is essential for their safety and functionality. Reinforcement learning from human feedback (RLHF) leverages human preferences to achieve this alignment. However, when preferences are sourced from diverse populations, point estimates of reward can result in suboptimal performance or be unfair to specific groups. We propose Pareto Optimal Preference Learning (POPL), which enables pluralistic alignment by framing discrepant group preferences as objectives with potential trade-offs, aiming for policies that are Pareto-optimal on the preference dataset. POPL utilizes lexicase selection, an iterative process that selects diverse and Pareto-optimal solutions. Our theoretical and empirical evaluations demonstrate that POPL surpasses baseline methods in learning sets of reward functions and policies, effectively catering to distinct groups without access to group numbers or membership labels. We verify the performance of POPL on a stateless preference learning setting, a Minigrid RL domain, Metaworld robotics benchmarks, as well as large language model (LLM) fine-tuning. We illustrate that POPL can also serve as a foundation for techniques optimizing specific notions of group fairness, ensuring safe and equitable AI model alignment.

Harshit Sikchi, Caleb Chuck, Amy Zhang et al.

Demonstrations are an effective alternative to task specification for learning agents in settings where designing a reward function is difficult. However, demonstrating expert behavior in the action space of the agent becomes unwieldy when robots have complex, unintuitive morphologies. We consider the practical setting where an agent has a dataset of prior interactions with the environment and is provided with observation-only expert demonstrations. Typical learning from observations approaches have required either learning an inverse dynamics model or a discriminator as intermediate steps of training. Errors in these intermediate one-step models compound during downstream policy learning or deployment. We overcome these limitations by directly learning a multi-step utility function that quantifies how each action impacts the agent's divergence from the expert's visitation distribution. Using the principle of duality, we derive DILO (Dual Imitation Learning from Observations), an algorithm that can leverage arbitrary suboptimal data to learn imitating policies without requiring expert actions. DILO reduces the learning from observations problem to that of simply learning an actor and a critic, bearing similar complexity to vanilla offline RL. This allows DILO to gracefully scale to high dimensional observations, and demonstrate improved performance across the board. Project page (code and videos): $\href{https://hari-sikchi.github.io/dilo/}{\text{hari-sikchi.github.io/dilo/}}$

Rafael Rafailov, Yaswanth Chittepu, Ryan Park et al.

Reinforcement Learning from Human Feedback (RLHF) has been crucial to the recent success of Large Language Models (LLMs), however, it is often a complex and brittle process. In the classical RLHF framework, a reward model is first trained to represent human preferences, which is in turn used by an online reinforcement learning (RL) algorithm to optimize the LLM. A prominent issue with such methods is reward over-optimization or reward hacking, where performance as measured by the learned proxy reward model increases, but true quality plateaus or even deteriorates. Direct Alignment Algorithms (DDAs) like Direct Preference Optimization have emerged as alternatives to the classical RLHF pipeline by circumventing the reward modeling phase. However, although DAAs do not use a separate proxy reward model, they still commonly deteriorate from over-optimization. While the so-called reward hacking phenomenon is not well-defined for DAAs, we still uncover similar trends: at higher KL budgets, DAA algorithms exhibit similar degradation patterns to their classic RLHF counterparts. In particular, we find that DAA methods deteriorate not only across a wide range of KL budgets but also often before even a single epoch of the dataset is completed. Through extensive empirical experimentation, this work formulates and formalizes the reward over-optimization or hacking problem for DAAs and explores its consequences across objectives, training regimes, and model scales.

Caleb Chuck, Carl Qi, Michael J. Munje et al.

Reinforcement Learning is a promising tool for learning complex policies even in fast-moving and object-interactive domains where human teleoperation or hard-coded policies might fail. To effectively reflect this challenging category of tasks, we introduce a dynamic, interactive RL testbed based on robot air hockey. By augmenting air hockey with a large family of tasks ranging from easy tasks like reaching, to challenging ones like pushing a block by hitting it with a puck, as well as goal-based and human-interactive tasks, our testbed allows a varied assessment of RL capabilities. The robot air hockey testbed also supports sim-to-real transfer with three domains: two simulators of increasing fidelity and a real robot system. Using a dataset of demonstration data gathered through two teleoperation systems: a virtualized control environment, and human shadowing, we assess the testbed with behavior cloning, offline RL, and RL from scratch.

Harshit Sikchi, Akanksha Saran, Wonjoon Goo et al.

We propose a new framework for imitation learning -- treating imitation as a two-player ranking-based game between a policy and a reward. In this game, the reward agent learns to satisfy pairwise performance rankings between behaviors, while the policy agent learns to maximize this reward. In imitation learning, near-optimal expert data can be difficult to obtain, and even in the limit of infinite data cannot imply a total ordering over trajectories as preferences can. On the other hand, learning from preferences alone is challenging as a large number of preferences are required to infer a high-dimensional reward function, though preference data is typically much easier to collect than expert demonstrations. The classical inverse reinforcement learning (IRL) formulation learns from expert demonstrations but provides no mechanism to incorporate learning from offline preferences and vice versa. We instantiate the proposed ranking-game framework with a novel ranking loss giving an algorithm that can simultaneously learn from expert demonstrations and preferences, gaining the advantages of both modalities. Our experiments show that the proposed method achieves state-of-the-art sample efficiency and can solve previously unsolvable tasks in the Learning from Observation (LfO) setting. Project video and code can be found at https://hari-sikchi.github.io/rank-game/

Christina J. Yuan, Yash Chandak, Stephen Giguere et al.

Many sequential decision making problems are high-stakes and require off-policy evaluation (OPE) of a new policy using historical data collected using some other policy. One of the most common OPE techniques that provides unbiased estimates is trajectory based importance sampling (IS). However, due to the high variance of trajectory IS estimates, importance sampling methods based on state-action visitation distributions (SIS) have recently been adopted. Unfortunately, while SIS often provides lower variance estimates for long horizons, estimating the state-action distribution ratios can be challenging and lead to biased estimates. In this paper, we present a new perspective on this bias-variance trade-off and show the existence of a spectrum of estimators whose endpoints are SIS and IS. Additionally, we also establish a spectrum for doubly-robust and weighted version of these estimators. We provide empirical evidence that estimators in this spectrum can be used to trade-off between the bias and variance of IS and SIS and can achieve lower mean-squared error than both IS and SIS.

Wonjoon Goo, Scott Niekum

The goal of offline reinforcement learning (RL) is to find an optimal policy given prerecorded trajectories. Many current approaches customize existing off-policy RL algorithms, especially actor-critic algorithms in which policy evaluation and improvement are iterated. However, the convergence of such approaches is not guaranteed due to the use of complex non-linear function approximation and an intertwined optimization process. By contrast, we propose a simple baseline algorithm for offline RL that only performs the policy evaluation step once so that the algorithm does not require complex stabilization schemes. Since the proposed algorithm is not likely to converge to an optimal policy, it is an appropriate baseline for actor-critic algorithms that ought to be outperformed if there is indeed value in iterative optimization in the offline setting. Surprisingly, we empirically find that the proposed algorithm exhibits competitive and sometimes even state-of-the-art performance in a subset of the D4RL offline RL benchmark. This result suggests that future work is needed to fully exploit the potential advantages of iterative optimization in order to justify the reduced stability of such methods.

Ajinkya Jain, Stephen Giguere, Rudolf Lioutikov et al.

We propose a method that efficiently learns distributions over articulation model parameters directly from depth images without the need to know articulation model categories a priori. By contrast, existing methods that learn articulation models from raw observations typically only predict point estimates of the model parameters, which are insufficient to guarantee the safe manipulation of articulated objects. Our core contributions include a novel representation for distributions over rigid body transformations and articulation model parameters based on screw theory, von Mises-Fisher distributions, and Stiefel manifolds. Combining these concepts allows for an efficient, mathematically sound representation that implicitly satisfies the constraints that rigid body transformations and articulations must adhere to. Leveraging this representation, we introduce a novel deep learning based approach, DUST-net, that performs category-independent articulation model estimation while also providing model uncertainties. We evaluate our approach on several benchmarking datasets and real-world objects and compare its performance with two current state-of-the-art methods. Our results demonstrate that DUST-net can successfully learn distributions over articulation models for novel objects across articulation model categories, which generate point estimates with better accuracy than state-of-the-art methods and effectively capture the uncertainty over predicted model parameters due to noisy inputs. Project webpage: https://pearl-utexas.github.io/DUST-net/

Farzan Memarian, Abolfazl Hashemi, Scott Niekum et al.

We explore methodologies to improve the robustness of generative adversarial imitation learning (GAIL) algorithms to observation noise. Towards this objective, we study the effect of local Lipschitzness of the discriminator and the generator on the robustness of policies learned by GAIL. In many robotics applications, the learned policies by GAIL typically suffer from a degraded performance at test time since the observations from the environment might be corrupted by noise. Hence, robustifying the learned policies against the observation noise is of critical importance. To this end, we propose a regularization method to induce local Lipschitzness in the generator and the discriminator of adversarial imitation learning methods. We show that the modified objective leads to learning significantly more robust policies. Moreover, we demonstrate -- both theoretically and experimentally -- that training a locally Lipschitz discriminator leads to a locally Lipschitz generator, thereby improving the robustness of the resultant policy. We perform extensive experiments on simulated robot locomotion environments from the MuJoCo suite that demonstrate the proposed method learns policies that significantly outperform the state-of-the-art generative adversarial imitation learning algorithm when applied to test scenarios with noise-corrupted observations.

Prasoon Goyal, Raymond J. Mooney, Scott Niekum

Imitation learning and instruction-following are two common approaches to communicate a user's intent to a learning agent. However, as the complexity of tasks grows, it could be beneficial to use both demonstrations and language to communicate with an agent. In this work, we propose a novel setting where an agent is given both a demonstration and a description, and must combine information from both the modalities. Specifically, given a demonstration for a task (the source task), and a natural language description of the differences between the demonstrated task and a related but different task (the target task), our goal is to train an agent to complete the target task in a zero-shot setting, that is, without any demonstrations for the target task. To this end, we introduce Language-Aided Reward and Value Adaptation (LARVA) which, given a source demonstration and a linguistic description of how the target task differs, learns to output a reward / value function that accurately describes the target task. Our experiments show that on a diverse set of adaptations, our approach is able to complete more than 95% of target tasks when using template-based descriptions, and more than 70% when using free-form natural language.

Ishan Durugkar, Mauricio Tec, Scott Niekum et al.

Learning with an objective to minimize the mismatch with a reference distribution has been shown to be useful for generative modeling and imitation learning. In this paper, we investigate whether one such objective, the Wasserstein-1 distance between a policy's state visitation distribution and a target distribution, can be utilized effectively for reinforcement learning (RL) tasks. Specifically, this paper focuses on goal-conditioned reinforcement learning where the idealized (unachievable) target distribution has full measure at the goal. This paper introduces a quasimetric specific to Markov Decision Processes (MDPs) and uses this quasimetric to estimate the above Wasserstein-1 distance. It further shows that the policy that minimizes this Wasserstein-1 distance is the policy that reaches the goal in as few steps as possible. Our approach, termed Adversarial Intrinsic Motivation (AIM), estimates this Wasserstein-1 distance through its dual objective and uses it to compute a supplemental reward function. Our experiments show that this reward function changes smoothly with respect to transitions in the MDP and directs the agent's exploration to find the goal efficiently. Additionally, we combine AIM with Hindsight Experience Replay (HER) and show that the resulting algorithm accelerates learning significantly on several simulated robotics tasks when compared to other rewards that encourage exploration or accelerate learning.

Yash Chandak, Scott Niekum, Bruno Castro da Silva et al.

When faced with sequential decision-making problems, it is often useful to be able to predict what would happen if decisions were made using a new policy. Those predictions must often be based on data collected under some previously used decision-making rule. Many previous methods enable such off-policy (or counterfactual) estimation of the expected value of a performance measure called the return. In this paper, we take the first steps towards a universal off-policy estimator (UnO) -- one that provides off-policy estimates and high-confidence bounds for any parameter of the return distribution. We use UnO for estimating and simultaneously bounding the mean, variance, quantiles/median, inter-quantile range, CVaR, and the entire cumulative distribution of returns. Finally, we also discuss Uno's applicability in various settings, including fully observable, partially observable (i.e., with unobserved confounders), Markovian, non-Markovian, stationary, smoothly non-stationary, and discrete distribution shifts.

Farzan Memarian, Wonjoon Goo, Rudolf Lioutikov et al.

We introduce Self-supervised Online Reward Shaping (SORS), which aims to improve the sample efficiency of any RL algorithm in sparse-reward environments by automatically densifying rewards. The proposed framework alternates between classification-based reward inference and policy update steps -- the original sparse reward provides a self-supervisory signal for reward inference by ranking trajectories that the agent observes, while the policy update is performed with the newly inferred, typically dense reward function. We introduce theory that shows that, under certain conditions, this alteration of the reward function will not change the optimal policy of the original MDP, while potentially increasing learning speed significantly. Experimental results on several sparse-reward environments demonstrate that, across multiple domains, the proposed algorithm is not only significantly more sample efficient than a standard RL baseline using sparse rewards, but, at times, also achieves similar sample efficiency compared to when hand-designed dense reward functions are used.

Mincheol Kim, Scott Niekum, Ashish D. Deshpande

We introduce a sample-efficient method for learning state-dependent stiffness control policies for dexterous manipulation. The ability to control stiffness facilitates safe and reliable manipulation by providing compliance and robustness to uncertainties. Most current reinforcement learning approaches to achieve robotic manipulation have exclusively focused on position control, often due to the difficulty of learning high-dimensional stiffness control policies. This difficulty can be partially mitigated via policy guidance such as imitation learning. However, expert stiffness control demonstrations are often expensive or infeasible to record. Therefore, we present an approach to learn Stiffness Control from Augmented Position control Experiences (SCAPE) that bypasses this difficulty by transforming position control demonstrations into approximate, suboptimal stiffness control demonstrations. Then, the suboptimality of the augmented demonstrations is addressed by using complementary techniques that help the agent safely learn from both the demonstrations and reinforcement learning. By using simulation tools and experiments on a robotic testbed, we show that the proposed approach efficiently learns safe manipulation policies and outperforms learned position control policies and several other baseline learning algorithms.

Daniel S. Brown, Jordan Schneider, Anca D. Dragan et al.

As humans interact with autonomous agents to perform increasingly complicated, potentially risky tasks, it is important to be able to efficiently evaluate an agent's performance and correctness. In this paper we formalize and theoretically analyze the problem of efficient value alignment verification: how to efficiently test whether the behavior of another agent is aligned with a human's values. The goal is to construct a kind of "driver's test" that a human can give to any agent which will verify value alignment via a minimal number of queries. We study alignment verification problems with both idealized humans that have an explicit reward function as well as problems where they have implicit values. We analyze verification of exact value alignment for rational agents and propose and analyze heuristic and approximate value alignment verification tests in a wide range of gridworlds and a continuous autonomous driving domain. Finally, we prove that there exist sufficient conditions such that we can verify exact and approximate alignment across an infinite set of test environments via a constant-query-complexity alignment test.

Yuchen Cui, Qiping Zhang, Alessandro Allievi et al.

Reactions such as gestures, facial expressions, and vocalizations are an abundant, naturally occurring channel of information that humans provide during interactions. A robot or other agent could leverage an understanding of such implicit human feedback to improve its task performance at no cost to the human. This approach contrasts with common agent teaching methods based on demonstrations, critiques, or other guidance that need to be attentively and intentionally provided. In this paper, we first define the general problem of learning from implicit human feedback and then propose to address this problem through a novel data-driven framework, EMPATHIC. This two-stage method consists of (1) mapping implicit human feedback to relevant task statistics such as reward, optimality, and advantage; and (2) using such a mapping to learn a task. We instantiate the first stage and three second-stage evaluations of the learned mapping. To do so, we collect a dataset of human facial reactions while participants observe an agent execute a sub-optimal policy for a prescribed training task. We train a deep neural network on this data and demonstrate its ability to (1) infer relative reward ranking of events in the training task from prerecorded human facial reactions; (2) improve the policy of an agent in the training task using live human facial reactions; and (3) transfer to a novel domain in which it evaluates robot manipulation trajectories.

Ajinkya Jain, Rudolf Lioutikov, Caleb Chuck et al.

Robots in human environments will need to interact with a wide variety of articulated objects such as cabinets, drawers, and dishwashers while assisting humans in performing day-to-day tasks. Existing methods either require objects to be textured or need to know the articulation model category a priori for estimating the model parameters for an articulated object. We propose ScrewNet, a novel approach that estimates an object's articulation model directly from depth images without requiring a priori knowledge of the articulation model category. ScrewNet uses screw theory to unify the representation of different articulation types and perform category-independent articulation model estimation. We evaluate our approach on two benchmarking datasets and compare its performance with a current state-of-the-art method. Results demonstrate that ScrewNet can successfully estimate the articulation models and their parameters for novel objects across articulation model categories with better on average accuracy than the prior state-of-the-art method. Project webpage: https://pearl-utexas.github.io/ScrewNet/

Prasoon Goyal, Scott Niekum, Raymond J. Mooney

Reinforcement learning (RL), particularly in sparse reward settings, often requires prohibitively large numbers of interactions with the environment, thereby limiting its applicability to complex problems. To address this, several prior approaches have used natural language to guide the agent's exploration. However, these approaches typically operate on structured representations of the environment, and/or assume some structure in the natural language commands. In this work, we propose a model that directly maps pixels to rewards, given a free-form natural language description of the task, which can then be used for policy learning. Our experiments on the Meta-World robot manipulation domain show that language-based rewards significantly improves the sample efficiency of policy learning, both in sparse and dense reward settings.

Akanksha Saran, Ruohan Zhang, Elaine Schaertl Short et al.

Human gaze is known to be an intention-revealing signal in human demonstrations of tasks. In this work, we use gaze cues from human demonstrators to enhance the performance of agents trained via three popular imitation learning methods -- behavioral cloning (BC), behavioral cloning from observation (BCO), and Trajectory-ranked Reward EXtrapolation (T-REX). Based on similarities between the attention of reinforcement learning agents and human gaze, we propose a novel approach for utilizing gaze data in a computationally efficient manner, as part of an auxiliary loss function, which guides a network to have higher activations in image regions where the human's gaze fixated. This work is a step towards augmenting any existing convolutional imitation learning agent's training with auxiliary gaze data. Our auxiliary coverage-based gaze loss (CGL) guides learning toward a better reward function or policy, without adding any additional learnable parameters and without requiring gaze data at test time. We find that our proposed approach improves the performance by 95% for BC, 343% for BCO, and 390% for T-REX, averaged over 20 different Atari games. We also find that compared to a prior state-of-the-art imitation learning method assisted by human gaze (AGIL), our method achieves better performance, and is more efficient in terms of learning with fewer demonstrations. We further interpret trained CGL agents with a saliency map visualization method to explain their performance. At last, we show that CGL can help alleviate a well-known causal confusion problem in imitation learning.