Jeff Schneider

Viraj Mehta, Ojash Neopane, Vikramjeet Das et al. · cmu

Preference-based feedback is important for many applications where direct evaluation of a reward function is not feasible. A notable recent example arises in reinforcement learning from human feedback on large language models. For many of these applications, the cost of acquiring the human feedback can be substantial or even prohibitive. In this work, we take advantage of the fact that often the agent can choose contexts at which to obtain human feedback in order to most efficiently identify a good policy, and introduce the offline contextual dueling bandit setting. We give an upper-confidence-bound style algorithm for this setting and prove a regret bound. We also give empirical confirmation that this method outperforms a similar strategy that uses uniformly sampled contexts.

Nikhil Angad Bakshi, Tejus Gupta, Ramina Ghods et al. · cmu

Robotic solutions for quick disaster response are essential to ensure minimal loss of life, especially when the search area is too dangerous or too vast for human rescuers. We model this problem as an asynchronous multi-agent active-search task where each robot aims to efficiently seek objects of interest (OOIs) in an unknown environment. This formulation addresses the requirement that search missions should focus on quick recovery of OOIs rather than full coverage of the search region. Previous approaches fail to accurately model sensing uncertainty, account for occlusions due to foliage or terrain, or consider the requirement for heterogeneous search teams and robustness to hardware and communication failures. We present the Generalized Uncertainty-aware Thompson Sampling (GUTS) algorithm, which addresses these issues and is suitable for deployment on heterogeneous multi-robot systems for active search in large unstructured environments. We show through simulation experiments that GUTS consistently outperforms existing methods such as parallelized Thompson Sampling and exhaustive search, recovering all OOIs in 80% of all runs. In contrast, existing approaches recover all OOIs in less than 40% of all runs. We conduct field tests using our multi-robot system in an unstructured environment with a search area of approximately 75,000 sq. m. Our system demonstrates robustness to various failure modes, achieving full recovery of OOIs (where feasible) in every field run, and significantly outperforming our baseline.

Viraj Mehta, Ian Char, Joseph Abbate et al.

Many potential applications of reinforcement learning (RL) are stymied by the large numbers of samples required to learn an effective policy. This is especially true when applying RL to real-world control tasks, e.g. in the sciences or robotics, where executing a policy in the environment is costly. In popular RL algorithms, agents typically explore either by adding stochasticity to a reward-maximizing policy or by attempting to gather maximal information about environment dynamics without taking the given task into account. In this work, we develop a method that allows us to plan for exploration while taking both the task and the current knowledge about the dynamics into account. The key insight to our approach is to plan an action sequence that maximizes the expected information gain about the optimal trajectory for the task at hand. We demonstrate that our method learns strong policies with 2x fewer samples than strong exploration baselines and 200x fewer samples than model free methods on a diverse set of low-to-medium dimensional control tasks in both the open-loop and closed-loop control settings.

Adam Villaflor, Zhe Huang, Swapnil Pande et al.

Impressive results in natural language processing (NLP) based on the Transformer neural network architecture have inspired researchers to explore viewing offline reinforcement learning (RL) as a generic sequence modeling problem. Recent works based on this paradigm have achieved state-of-the-art results in several of the mostly deterministic offline Atari and D4RL benchmarks. However, because these methods jointly model the states and actions as a single sequencing problem, they struggle to disentangle the effects of the policy and world dynamics on the return. Thus, in adversarial or stochastic environments, these methods lead to overly optimistic behavior that can be dangerous in safety-critical systems like autonomous driving. In this work, we propose a method that addresses this optimism bias by explicitly disentangling the policy and world models, which allows us at test time to search for policies that are robust to multiple possible futures in the environment. We demonstrate our method's superior performance on a variety of autonomous driving tasks in simulation.

Conor Igoe, Youngseog Chung, Ian Char et al.

One critical challenge in deploying highly performant machine learning models in real-life applications is out of distribution (OOD) detection. Given a predictive model which is accurate on in distribution (ID) data, an OOD detection system will further equip the model with the option to defer prediction when the input is novel and the model has little confidence in prediction. There has been some recent interest in utilizing the gradient information in pre-trained models for OOD detection. While these methods have shown competitive performance, there are misconceptions about the true mechanism underlying them, which conflate their performance with the necessity of gradients. In this work, we provide an in-depth analysis and comparison of gradient based methods and elucidate the key components that warrant their OOD detection performance. We further propose a general, non-gradient based method of OOD detection which improves over previous baselines in both performance and computational efficiency.

Aditya Kapoor, Benjamin Freed, Howie Choset et al.

Multi-agent proximal policy optimization (MAPPO) has recently demonstrated state-of-the-art performance on challenging multi-agent reinforcement learning tasks. However, MAPPO still struggles with the credit assignment problem, wherein the sheer difficulty in ascribing credit to individual agents' actions scales poorly with team size. In this paper, we propose a multi-agent reinforcement learning algorithm that adapts recent developments in credit assignment to improve upon MAPPO. Our approach leverages partial reward decoupling (PRD), which uses a learned attention mechanism to estimate which of a particular agent's teammates are relevant to its learning updates. We use this estimate to dynamically decompose large groups of agents into smaller, more manageable subgroups. We empirically demonstrate that our approach, PRD-MAPPO, decouples agents from teammates that do not influence their expected future reward, thereby streamlining credit assignment. We additionally show that PRD-MAPPO yields significantly higher data efficiency and asymptotic performance compared to both MAPPO and other state-of-the-art methods across several multi-agent tasks, including StarCraft II. Finally, we propose a version of PRD-MAPPO that is applicable to \textit{shared} reward settings, where PRD was previously not applicable, and empirically show that this also leads to performance improvements over MAPPO.

Xiang Li, Viraj Mehta, Johannes Kirschner et al.

Many real-world reinforcement learning tasks require control of complex dynamical systems that involve both costly data acquisition processes and large state spaces. In cases where the transition dynamics can be readily evaluated at specified states (e.g., via a simulator), agents can operate in what is often referred to as planning with a \emph{generative model}. We propose the AE-LSVI algorithm for best-policy identification, a novel variant of the kernelized least-squares value iteration (LSVI) algorithm that combines optimism with pessimism for active exploration (AE). AE-LSVI provably identifies a near-optimal policy \emph{uniformly} over an entire state space and achieves polynomial sample complexity guarantees that are independent of the number of states. When specialized to the recently introduced offline contextual Bayesian optimization setting, our algorithm achieves improved sample complexity bounds. Experimentally, we demonstrate that AE-LSVI outperforms other RL algorithms in a variety of environments when robustness to the initial state is required.

Ian Char, Viraj Mehta, Adam Villaflor et al.

The problem of offline reinforcement learning focuses on learning a good policy from a log of environment interactions. Past efforts for developing algorithms in this area have revolved around introducing constraints to online reinforcement learning algorithms to ensure the actions of the learned policy are constrained to the logged data. In this work, we explore an alternative approach by planning on the fixed dataset directly. Specifically, we introduce an algorithm which forms a tabular Markov Decision Process (MDP) over the logged data by adding new transitions to the dataset. We do this by using learned dynamics models to plan short trajectories between states. Since exact value iteration can be performed on this constructed MDP, it becomes easy to identify which trajectories are advantageous to add to the MDP. Crucially, since most transitions in this MDP come from the logged data, trajectories from the MDP can be rolled out for long periods with confidence. We prove that this property allows one to make upper and lower bounds on the value function up to appropriate distance metrics. Finally, we demonstrate empirically how algorithms that uniformly constrain the learned policy to the entire dataset can result in unwanted behavior, and we show an example in which simply behavior cloning the optimal policy of the MDP created by our algorithm avoids this problem.

Ian Char, Jeff Schneider

Deep reinforcement learning (RL) has shown immense potential for learning to control systems through data alone. However, one challenge deep RL faces is that the full state of the system is often not observable. When this is the case, the policy needs to leverage the history of observations to infer the current state. At the same time, differences between the training and testing environments makes it critical for the policy not to overfit to the sequence of observations it sees at training time. As such, there is an important balancing act between having the history encoder be flexible enough to extract relevant information, yet be robust to changes in the environment. To strike this balance, we look to the PID controller for inspiration. We assert the PID controller's success shows that only summing and differencing are needed to accumulate information over time for many control tasks. Following this principle, we propose two architectures for encoding history: one that directly uses PID features and another that extends these core ideas and can be used in arbitrary control tasks. When compared with prior approaches, our encoders produce policies that are often more robust and achieve better performance on a variety of tracking tasks. Going beyond tracking tasks, our policies achieve 1.7x better performance on average over previous state-of-the-art methods on a suite of locomotion control tasks.

Arundhati Banerjee, Ramina Ghods, Jeff Schneider

Active search, in applications like environment monitoring or disaster response missions, involves autonomous agents detecting targets in a search space using decision making algorithms that adapt to the history of their observations. Active search algorithms must contend with two types of uncertainty: detection uncertainty and location uncertainty. The more common approach in robotics is to focus on location uncertainty and remove detection uncertainty by thresholding the detection probability to zero or one. In contrast, it is common in the sparse signal processing literature to assume the target location is accurate and instead focus on the uncertainty of its detection. In this work, we first propose an inference method to jointly handle both target detection and location uncertainty. We then build a decision making algorithm on this inference method that uses Thompson sampling to enable decentralized multi-agent active search. We perform simulation experiments to show that our algorithms outperform competing baselines that only account for either target detection or location uncertainty. We finally demonstrate the real world transferability of our algorithms using a realistic simulation environment we created on the Unreal Engine 4 platform with an AirSim plugin.

Vikram Duvvur, Aashay Mehta, Edward Sun et al.

The use of machine learning in algorithmic trading systems is increasingly common. In a typical set-up, supervised learning is used to predict the future prices of assets, and those predictions drive a simple trading and execution strategy. This is quite effective when the predictions have sufficient signal, markets are liquid, and transaction costs are low. However, those conditions often do not hold in thinly traded financial markets and markets for differentiated assets such as real estate or vehicles. In these markets, the trading strategy must consider the long-term effects of taking positions that are relatively more difficult to change. In this work, we propose a Reinforcement Learning (RL) algorithm that trades based on signals from a learned predictive model and addresses these challenges. We test our algorithm on 20+ years of equity data from Bursa Malaysia.

Anirudha Ramesh, Anurag Ghosh, Christoph Mertz et al.

Robotic Perception in diverse domains such as low-light scenarios, where new modalities like thermal imaging and specialized night-vision sensors are increasingly employed, remains a challenge. Largely, this is due to the limited availability of labeled data. Existing Domain Adaptation (DA) techniques, while promising to leverage labels from existing well-lit RGB images, fail to consider the characteristics of the source domain itself. We holistically account for this factor by proposing Source Preparation (SP), a method to mitigate source domain biases. Our Almost Unsupervised Domain Adaptation (AUDA) framework, a label-efficient semi-supervised approach for robotic scenarios -- employs Source Preparation (SP), Unsupervised Domain Adaptation (UDA) and Supervised Alignment (SA) from limited labeled data. We introduce CityIntensified, a novel dataset comprising temporally aligned image pairs captured from a high-sensitivity camera and an intensifier camera for semantic segmentation and object detection in low-light settings. We demonstrate the effectiveness of our method in semantic segmentation, with experiments showing that SP enhances UDA across a range of visual domains, with improvements up to 40.64% in mIoU over baseline, while making target models more robust to real-world shifts within the target domain. We show that AUDA is a label-efficient framework for effective DA, significantly improving target domain performance with only tens of labeled samples from the target domain.

Arundhati Banerjee, Ramina Ghods, Jeff Schneider

Multi-agent active search requires autonomous agents to choose sensing actions that efficiently locate targets. In a realistic setting, agents also must consider the costs that their decisions incur. Previously proposed active search algorithms simplify the problem by ignoring uncertainty in the agent's environment, using myopic decision making, and/or overlooking costs. In this paper, we introduce an online active search algorithm to detect targets in an unknown environment by making adaptive cost-aware decisions regarding the agent's actions. Our algorithm combines principles from Thompson Sampling (for search space exploration and decentralized multi-agent decision making), Monte Carlo Tree Search (for long horizon planning) and pareto-optimal confidence bounds (for multi-objective optimization in an unknown environment) to propose an online lookahead planner that removes all the simplifications. We analyze the algorithm's performance in simulation to show its efficacy in cost aware active search.

Siddarth Venkatraman, Shivesh Khaitan, Ravi Tej Akella et al.

Offline reinforcement learning (RL) holds promise as a means to learn high-reward policies from a static dataset, without the need for further environment interactions. However, a key challenge in offline RL lies in effectively stitching portions of suboptimal trajectories from the static dataset while avoiding extrapolation errors arising due to a lack of support in the dataset. Existing approaches use conservative methods that are tricky to tune and struggle with multi-modal data (as we show) or rely on noisy Monte Carlo return-to-go samples for reward conditioning. In this work, we propose a novel approach that leverages the expressiveness of latent diffusion to model in-support trajectory sequences as compressed latent skills. This facilitates learning a Q-function while avoiding extrapolation error via batch-constraining. The latent space is also expressive and gracefully copes with multi-modal data. We show that the learned temporally-abstract latent space encodes richer task-specific information for offline RL tasks as compared to raw state-actions. This improves credit assignment and facilitates faster reward propagation during Q-learning. Our method demonstrates state-of-the-art performance on the D4RL benchmarks, particularly excelling in long-horizon, sparse-reward tasks.

Zhouchonghao Wu, Raymond Song, Vedant Mundheda et al.

Off-road autonomous driving poses significant challenges such as navigating unmapped, variable terrain with uncertain and diverse dynamics. Addressing these challenges requires effective long-horizon planning and adaptable control. Reinforcement Learning (RL) offers a promising solution by learning control policies directly from interaction. However, because off-road driving is a long-horizon task with low-signal rewards, standard RL methods are challenging to apply in this setting. We introduce TADPO, a novel policy gradient formulation that extends Proximal Policy Optimization (PPO), leveraging off-policy trajectories for teacher guidance and on-policy trajectories for student exploration. Building on this, we develop a vision-based, end-to-end RL system for high-speed off-road driving, capable of navigating extreme slopes and obstacle-rich terrain. We demonstrate our performance in simulation and, importantly, zero-shot sim-to-real transfer on a full-scale off-road vehicle. To our knowledge, this work represents the first deployment of RL-based policies on a full-scale off-road platform.

Jiayu Chen, Le Xu, Wentse Chen et al.

Offline RL is a powerful approach for data-driven decision-making and control. Compared to model-free methods, offline model-based RL (MBRL) explicitly learns world models from a static dataset and uses them as surrogate simulators, improving the data efficiency and enabling the learned policy to potentially generalize beyond the dataset support. However, there could be various MDPs that behave identically on the offline dataset and dealing with the uncertainty about the true MDP can be challenging. In this paper, we propose modeling offline MBRL as a Bayes Adaptive Markov Decision Process (BAMDP), which is a principled framework for addressing model uncertainty. We further propose a novel Bayes Adaptive Monte-Carlo planning algorithm capable of solving BAMDPs in continuous state and action spaces with stochastic transitions. This planning process is based on Monte Carlo Tree Search and can be integrated into offline MBRL as a policy improvement operator in policy iteration. Our ``RL + Search" framework follows in the footsteps of superhuman AIs like AlphaZero, improving on current offline MBRL methods by incorporating more computation input. The proposed algorithm significantly outperforms state-of-the-art offline RL methods on twelve D4RL MuJoCo tasks and three target tracking tasks in a challenging, stochastic tokamak control simulator. The codebase is available at: https://github.com/LucasCJYSDL/Offline-RL-Kit.

Aravind Venugopal, Jiayu Chen, Xudong Wu et al.

The temporal lag between actions and their long-term consequences makes credit assignment a challenge when learning goal-directed behaviors from data. Generative world models capture the distribution of future states an agent may visit, indicating that they have captured temporal information. How can that temporal information be extracted to perform credit assignment? In this paper, we formalize how the temporal information stored in world models encodes the underlying geometry of the world. Leveraging optimal transport, we extract this geometry from a learned model of the occupancy measure into a reward function that captures goal-reaching information. Our resulting method, Occupancy Reward Shaping, largely mitigates the problem of credit assignment in sparse reward settings. ORS provably does not alter the optimal policy, yet empirically improves performance by 2.2x across 13 diverse long-horizon locomotion and manipulation tasks. Moreover, we demonstrate the effectiveness of ORS in the real world for controlling nuclear fusion on 3 Tokamak control tasks. Code: https://github.com/aravindvenu7/occupancy_reward_shaping; Website: https://aravindvenu7.github.io/website/ors/

Youngseog Chung, Dhruv Malik, Jeff Schneider et al.

The traditional viewpoint on Sparse Mixture of Experts (MoE) models is that instead of training a single large expert, which is computationally expensive, we can train many small experts. The hope is that if the total parameter count of the small experts equals that of the singular large expert, then we retain the representation power of the large expert while gaining computational tractability and promoting expert specialization. The recently introduced Soft MoE replaces the Sparse MoE's discrete routing mechanism with a differentiable gating function that smoothly mixes tokens. While this smooth gating function successfully mitigates the various training instabilities associated with Sparse MoE, it is unclear whether it induces implicit biases that affect Soft MoE's representation power or potential for expert specialization. We prove that Soft MoE with a single arbitrarily powerful expert cannot represent simple convex functions. This justifies that Soft MoE's success cannot be explained by the traditional viewpoint of many small experts collectively mimicking the representation power of a single large expert, and that multiple experts are actually necessary to achieve good representation power (even for a fixed total parameter count). Continuing along this line of investigation, we introduce a notion of expert specialization for Soft MoE, and while varying the number of experts yet fixing the total parameter count, we consider the following (computationally intractable) task. Given any input, how can we discover the expert subset that is specialized to predict this input's label? We empirically show that when there are many small experts, the architecture is implicitly biased in a fashion that allows us to efficiently approximate the specialized expert subset. Our method can be easily implemented to potentially reduce computation during inference.

Yuxuan Li, Qijun He, Mingqi Yuan et al.

Continual Reinforcement Learning (CRL) aims to develop lifelong learning agents to continuously acquire knowledge across diverse tasks while mitigating catastrophic forgetting. This requires efficiently managing the stability-plasticity dilemma and leveraging prior experience to rapidly generalize to novel tasks. While various enhancement strategies for both aspects have been proposed, achieving scalable performance by directly applying RL to sequential task streams remains challenging. In this paper, we propose a novel teacher-student framework that decouples CRL into two independent processes: training single-task teacher models through distributed RL and continually distilling them into a central generalist model. This design is motivated by the observation that RL excels at solving single tasks, while policy distillation -- a relatively stable supervised learning process -- is well aligned with large foundation models and multi-task learning. Moreover, a mixture-of-experts (MoE) architecture and a replay-based approach are employed to enhance the plasticity and stability of the continual policy distillation process. Extensive experiments on the Meta-World benchmark demonstrate that our framework enables efficient continual RL, recovering over 85% of teacher performance while constraining task-wise forgetting to within 10%.

Wen-Tse Chen, Jiayu Chen, Fahim Tajwar et al.

Learning from self-sampled data and sparse environmental feedback remains a fundamental challenge in training self-evolving agents. Temporal credit assignment mitigates this issue by transforming sparse feedback into dense supervision signals. However, previous approaches typically depend on learning task-specific value functions for credit assignment, which suffer from poor sample efficiency and limited generalization. In this work, we propose to leverage pretrained knowledge from large language models (LLMs) to transform sparse rewards into dense training signals (i.e., the advantage function) through retrospective in-context learning (RICL). We further propose an online learning framework, RICOL, which iteratively refines the policy based on the credit assignment results from RICL. We empirically demonstrate that RICL can accurately estimate the advantage function with limited samples and effectively identify critical states in the environment for temporal credit assignment. Extended evaluation on four BabyAI scenarios show that RICOL achieves comparable convergent performance with traditional online RL algorithms with significantly higher sample efficiency. Our findings highlight the potential of leveraging LLMs for temporal credit assignment, paving the way for more sample-efficient and generalizable RL paradigms.

Fahim Tajwar, Anikait Singh, Archit Sharma et al. · stanford

Learning from preference labels plays a crucial role in fine-tuning large language models. There are several distinct approaches for preference fine-tuning, including supervised learning, on-policy reinforcement learning (RL), and contrastive learning. Different methods come with different implementation tradeoffs and performance differences, and existing empirical findings present different conclusions, for instance, some results show that online RL is quite important to attain good fine-tuning results, while others find (offline) contrastive or even purely supervised methods sufficient. This raises a natural question: what kind of approaches are important for fine-tuning with preference data and why? In this paper, we answer this question by performing a rigorous analysis of a number of fine-tuning techniques on didactic and full-scale LLM problems. Our main finding is that, in general, approaches that use on-policy sampling or attempt to push down the likelihood on certain responses (i.e., employ a "negative gradient") outperform offline and maximum likelihood objectives. We conceptualize our insights and unify methods that use on-policy sampling or negative gradient under a notion of mode-seeking objectives for categorical distributions. Mode-seeking objectives are able to alter probability mass on specific bins of a categorical distribution at a fast rate compared to maximum likelihood, allowing them to relocate masses across bins more effectively. Our analysis prescribes actionable insights for preference fine-tuning of LLMs and informs how data should be collected for maximal improvement.

Youngseog Chung, Ian Char, Han Guo et al.

With increasing deployment of machine learning systems in various real-world tasks, there is a greater need for accurate quantification of predictive uncertainty. While the common goal in uncertainty quantification (UQ) in machine learning is to approximate the true distribution of the target data, many works in UQ tend to be disjoint in the evaluation metrics utilized, and disparate implementations for each metric lead to numerical results that are not directly comparable across different works. To address this, we introduce Uncertainty Toolbox, an open-source python library that helps to assess, visualize, and improve UQ. Uncertainty Toolbox additionally provides pedagogical resources, such as a glossary of key terms and an organized collection of key paper references. We hope that this toolbox is useful for accelerating and uniting research efforts in uncertainty in machine learning.

Kirthevasan Kandasamy, Karun Raju Vysyaraju, Willie Neiswanger et al.

Bayesian Optimisation (BO) refers to a suite of techniques for global optimisation of expensive black box functions, which use introspective Bayesian models of the function to efficiently search for the optimum. While BO has been applied successfully in many applications, modern optimisation tasks usher in new challenges where conventional methods fail spectacularly. In this work, we present Dragonfly, an open source Python library for scalable and robust BO. Dragonfly incorporates multiple recently developed methods that allow BO to be applied in challenging real world settings; these include better methods for handling higher dimensional domains, methods for handling multi-fidelity evaluations when cheap approximations of an expensive function are available, methods for optimising over structured combinatorial spaces, such as the space of neural network architectures, and methods for handling parallel evaluations. Additionally, we develop new methodological improvements in BO for selecting the Bayesian model, selecting the acquisition function, and optimising over complex domains with different variable types and additional constraints. We compare Dragonfly to a suite of other packages and algorithms for global optimisation and demonstrate that when the above methods are integrated, they enable significant improvements in the performance of BO. The Dragonfly library is available at dragonfly.github.io.

Rohit Sonker, Hiro Josep Farre Kaga, Jiayu Chen et al.

Tokamaks remain leading candidates for achieving practical fusion energy, yet many important control problems inside these devices are still difficult or unsolved. One such challenge is controlling the plasma rotation profile, which strongly influences stability, confinement, and transport. While the average rotation can be controlled, controlling the full profile is challenging due to high dimensionality, response to multiple actuators and dependence on plasma condition. Learning-based control methods, such as reinforcement learning (RL), provide a potential solution to this challenging problem with ability to model complex interactions leading to effective multi-input multi-output control. However, learning such policies is challenging due to the lack of accurate simulators that can model the rotation profile dynamics. In this work, we investigate the use of offline RL and offline model-based RL algorithms for rotation profile control, training them solely on historical data from the DIII-D tokamak. Our final method uses probabilistic models of plasma dynamics to generate rollouts for RL training. We deploy this policy on the DIII-D Tokamak and observe promising real-world results. We conclude by highlighting key challenges and insights from training and deploying an RL policy on a complex physical device while using only limited past data.

Sang Keun Choe, Hwijeen Ahn, Juhan Bae et al. · cmu, utoronto

Large language models (LLMs) are trained on a vast amount of human-written data, but data providers often remain uncredited. In response to this issue, data valuation (or data attribution), which quantifies the contribution or value of each data to the model output, has been discussed as a potential solution. Nevertheless, applying existing data valuation methods to recent LLMs and their vast training datasets has been largely limited by prohibitive compute and memory costs. In this work, we focus on influence functions, a popular gradient-based data valuation method, and significantly improve its scalability with an efficient gradient projection strategy called LoGra that leverages the gradient structure in backpropagation. We then provide a theoretical motivation of gradient projection approaches to influence functions to promote trust in the data valuation process. Lastly, we lower the barrier to implementing data valuation systems by introducing LogIX, a software package that can transform existing training code into data valuation code with minimal effort. In our data valuation experiments, LoGra achieves competitive accuracy against more expensive baselines while showing up to 6,500x improvement in throughput and 5x reduction in GPU memory usage when applied to Llama3-8B-Instruct and the 1B-token dataset.

Brian Yang, Huangyuan Su, Nikolaos Gkanatsios et al.

Diffusion models excel at modeling complex and multimodal trajectory distributions for decision-making and control. Reward-gradient guided denoising has been recently proposed to generate trajectories that maximize both a differentiable reward function and the likelihood under the data distribution captured by a diffusion model. Reward-gradient guided denoising requires a differentiable reward function fitted to both clean and noised samples, limiting its applicability as a general trajectory optimizer. In this paper, we propose DiffusionES, a method that combines gradient-free optimization with trajectory denoising to optimize black-box non-differentiable objectives while staying in the data manifold. Diffusion-ES samples trajectories during evolutionary search from a diffusion model and scores them using a black-box reward function. It mutates high-scoring trajectories using a truncated diffusion process that applies a small number of noising and denoising steps, allowing for much more efficient exploration of the solution space. We show that DiffusionES achieves state-of-the-art performance on nuPlan, an established closed-loop planning benchmark for autonomous driving. Diffusion-ES outperforms existing sampling-based planners, reactive deterministic or diffusion-based policies, and reward-gradient guidance. Additionally, we show that unlike prior guidance methods, our method can optimize non-differentiable language-shaped reward functions generated by few-shot LLM prompting. When guided by a human teacher that issues instructions to follow, our method can generate novel, highly complex behaviors, such as aggressive lane weaving, which are not present in the training data. This allows us to solve the hardest nuPlan scenarios which are beyond the capabilities of existing trajectory optimization methods and driving policies.

Sheikh Shafayat, Fahim Tajwar, Ruslan Salakhutdinov et al.

Recent successes of reinforcement learning (RL) in training large reasoning models motivate the question of whether self-training - the process where a model learns from its own judgments - can be sustained within RL. In this work, we study this question using majority voting as a simple self-feedback mechanism. On a comprehensive set of experiments on both synthetic and real reasoning tasks, we find that this basic approach improves not only the model's reasoning performance, but also its capability of generating better quality feedback for the next RL iteration, driving further model improvement. Yet our analysis also reveals a critical limitation of such a self-training paradigm - prolonged RL with self-reward leads to reward hacking where models learn to maximize training (pseudo-)reward, resulting in sudden and complete performance collapse. Together, these results highlight feedback design as the central challenge and call for future research on mechanisms to enable prolonged self-improvement.

Ian Char, Youngseog Chung, Joseph Abbate et al.

Although tokamaks are one of the most promising devices for realizing nuclear fusion as an energy source, there are still key obstacles when it comes to understanding the dynamics of the plasma and controlling it. As such, it is crucial that high quality models are developed to assist in overcoming these obstacles. In this work, we take an entirely data driven approach to learn such a model. In particular, we use historical data from the DIII-D tokamak to train a deep recurrent network that is able to predict the full time evolution of plasma discharges (or "shots"). Following this, we investigate how different training and inference procedures affect the quality and calibration of the shot predictions.

Arundhati Banerjee, Jeff Schneider

Multi-agent multi-target tracking has a wide range of applications, including wildlife patrolling, security surveillance or environment monitoring. Such algorithms often make restrictive assumptions: the number of targets and/or their initial locations may be assumed known, or agents may be pre-assigned to monitor disjoint partitions of the environment, reducing the burden of exploration. This also limits applicability when there are fewer agents than targets, since agents are unable to continuously follow the targets in their fields of view. Multi-agent tracking algorithms additionally assume inter-agent synchronization of observations, or the presence of a central controller to coordinate joint actions. Instead, we focus on the setting of decentralized multi-agent, multi-target, simultaneous active search-and-tracking with asynchronous inter-agent communication. Our proposed algorithm DecSTER uses a sequential monte carlo implementation of the probability hypothesis density filter for posterior inference combined with Thompson sampling for decentralized multi-agent decision making. We compare different action selection policies, focusing on scenarios where targets outnumber agents. In simulation, we demonstrate that DecSTER is robust to unreliable inter-agent communication and outperforms information-greedy baselines in terms of the Optimal Sub-Pattern Assignment (OSPA) metric for different numbers of targets and varying teamsizes.

Fahim Tajwar, Yiding Jiang, Abitha Thankaraj et al.

Efficient exploration is essential for intelligent systems interacting with their environment, but existing language models often fall short in scenarios that require strategic information gathering. In this paper, we present Paprika, a fine-tuning approach that enables language models to develop general decision-making capabilities that are not confined to particular environments. By training on synthetic interaction data from different tasks that require diverse strategies, Paprika teaches models to explore and adapt their behavior on a new task based on environment feedback in-context without more gradient updates. Experimental results show that models fine-tuned with Paprika can effectively transfer their learned decision-making capabilities to entirely unseen tasks without additional training. Unlike traditional training, our approach's primary bottleneck lies in sampling useful interaction data instead of model updates. To improve sample efficiency, we propose a curriculum learning strategy that prioritizes sampling trajectories from tasks with high learning potential. These results suggest a promising path towards AI systems that can autonomously solve novel sequential decision-making problems that require interactions with the external world.

Arundhati Banerjee, Jeff Schneider

Active search for recovering objects of interest through online, adaptive decision making with autonomous agents requires trading off exploration of unknown environments with exploitation of prior observations in the search space. Prior work has proposed information gain and Thompson sampling based myopic, greedy approaches for agents to actively decide query or search locations when the number of targets is unknown. Decision making algorithms in such partially observable environments have also shown that agents capable of lookahead over a finite horizon outperform myopic policies for active search. Unfortunately, lookahead algorithms typically rely on building a computationally expensive search tree that is simulated and updated based on the agent's observations and a model of the environment dynamics. Instead, in this work, we leverage the sequence modeling abilities of diffusion models to sample lookahead action sequences that balance the exploration-exploitation trade-off for active search without building an exhaustive search tree. We identify the optimism bias in prior diffusion based reinforcement learning approaches when applied to the active search setting and propose mitigating solutions for efficient cost-aware decision making with both single and multi-agent teams. Our proposed algorithm outperforms standard baselines in offline reinforcement learning in terms of full recovery rate and is computationally more efficient than tree search in cost-aware active decision making.

Haotian Lin, Pengcheng Wang, Jeff Schneider et al.

Model-based reinforcement learning algorithms that combine model-based planning and learned value/policy prior have gained significant recognition for their high data efficiency and superior performance in continuous control. However, we discover that existing methods that rely on standard SAC-style policy iteration for value learning, directly using data generated by the planner, often result in \emph{persistent value overestimation}. Through theoretical analysis and experiments, we argue that this issue is deeply rooted in the structural policy mismatch between the data generation policy that is always bootstrapped by the planner and the learned policy prior. To mitigate such a mismatch in a minimalist way, we propose a policy regularization term reducing out-of-distribution (OOD) queries, thereby improving value learning. Our method involves minimum changes on top of existing frameworks and requires no additional computation. Extensive experiments demonstrate that the proposed approach improves performance over baselines such as TD-MPC2 by large margins, particularly in 61-DoF humanoid tasks. View qualitative results at https://darthutopian.github.io/tdmpc_square/.

Fahim Tajwar, Guanning Zeng, Yueer Zhou et al.

Reinforcement learning is the method of choice to train models in sampling-based setups with binary outcome feedback, such as navigation, code generation, and mathematical problem solving. In such settings, models implicitly induce a likelihood over correct rollouts. However, we observe that reinforcement learning does not maximize this likelihood, and instead optimizes only a lower-order approximation. Inspired by this observation, we introduce Maximum Likelihood Reinforcement Learning (MaxRL), a sampling-based framework to approximate maximum likelihood using reinforcement learning techniques. MaxRL addresses the challenges of non-differentiable sampling by defining a compute-indexed family of sample-based objectives that interpolate between standard reinforcement learning and exact maximum likelihood as additional sampling compute is allocated. The resulting objectives admit a simple, unbiased policy-gradient estimator and converge to maximum likelihood optimization in the infinite-compute limit. Empirically, we show that MaxRL Pareto-dominates existing methods in all models and tasks we tested, achieving up to 20x test-time scaling efficiency gains compared to its GRPO-trained counterpart. We also observe MaxRL to scale better with additional data and compute. Our results suggest MaxRL is a promising framework for scaling RL training in correctness based settings.

Yiqi Wang, Mrinal Verghese, Jeff Schneider

Learning visuomotor policies via imitation has proven effective across a wide range of robotic domains. However, the performance of these policies is heavily dependent on the number of training demonstrations, which requires expensive data collection in the real world. In this work, we aim to reduce data collection efforts when learning visuomotor robot policies by leveraging existing or cost-effective data from a wide range of embodiments, such as public robot datasets and the datasets of humans playing with objects (human data from play). Our approach leverages two key insights. First, we use optic flow as an embodiment-agnostic action representation to train a World Model (WM) across multi-embodiment datasets, and finetune it on a small amount of robot data from the target embodiment. Second, we develop a method, Latent Policy Steering (LPS), to improve the output of a behavior-cloned policy by searching in the latent space of the WM for better action sequences. In real world experiments, we observe significant improvements in the performance of policies trained with a small amount of data (over 50% relative improvement with 30 demonstrations and over 20% relative improvement with 50 demonstrations) by combining the policy with a WM pretrained on two thousand episodes sampled from the existing Open X-embodiment dataset across different robots or a cost-effective human dataset from play.

Xintong Duan, Yutong He, Fahim Tajwar et al.

Although diffusion models have achieved strong results in decision-making tasks, their slow inference speed remains a key limitation. While the consistency model offers a potential solution, its applications to decision-making often struggle with suboptimal demonstrations or rely on complex concurrent training of multiple networks. In this work, we propose a novel approach to consistency distillation for offline reinforcement learning that directly incorporates reward optimization into the distillation process. Our method enables single-step generation while maintaining higher performance and simpler training. Empirical evaluations on the Gym MuJoCo benchmarks and long horizon planning demonstrate that our approach can achieve an 8.7% improvement over previous state-of-the-art while offering up to 142x speedup over diffusion counterparts in inference time.

Xintong Duan, Yutong He, Fahim Tajwar et al.

Many real-world decision-making problems are combinatorial in nature, where states (e.g., surrounding traffic of a self-driving car) can be seen as a combination of basic elements (e.g., pedestrians, trees, and other cars). Due to combinatorial complexity, observing all combinations of basic elements in the training set is infeasible, which leads to an essential yet understudied problem of zero-shot generalization to states that are unseen combinations of previously seen elements. In this work, we first formalize this problem and then demonstrate how existing value-based reinforcement learning (RL) algorithms struggle due to unreliable value predictions in unseen states. We argue that this problem cannot be addressed with exploration alone, but requires more expressive and generalizable models. We demonstrate that behavior cloning with a conditioned diffusion model trained on expert trajectory generalizes better to states formed by new combinations of seen elements than traditional RL methods. Through experiments in maze, driving, and multiagent environments, we show that conditioned diffusion models outperform traditional RL techniques and highlight the broad applicability of our problem formulation.

Grace Liu, Yuxiao Qu, Jeff Schneider et al.

Many tasks require learned models to strategically gather relevant information over multiple rounds of interaction before actually acting on a task. Strategic information gathering requires models to know not only how to effectively acquire information, but also when to stop gathering information and make a decision, in order to avoid overthinking or getting derailed when acting. In this paper, we formalize this problem and introduce Counterfactuals and Reasoning for Termination (CaRT), an approach for teaching LLMs when to stop seeking information. To appropriately learn when to terminate, CaRT fine-tunes LLMs using counterfactual pairs of trajectories, one where termination is appropriate and a minimally modified version of the same trajectory where it is not. It trains the LLM to explain the rationale for the termination decision in either case via verbal reasoning, and imbues this capability into the base LLM via fine-tuning. We instantiate CaRT in two domains: interactive medical diagnosis and math problem solving. In both domains, we find that CaRT improves the efficiency of information gathering and task success rate compared to other fine-tuning methods.

Aman Mehra, Alexandre Capone, Jeff Schneider

A long-standing problem in online reinforcement learning (RL) is of ensuring sample efficiency, which stems from an inability to explore environments efficiently. Most attempts at efficient exploration tackle this problem in a setting where learning begins from scratch, without prior information available to bootstrap learning. However, such approaches fail to leverage expert demonstrations and simulators that can reset to arbitrary states. These affordances are valuable resources that offer enormous potential to guide exploration and speed up learning. In this paper, we explore how a small number of expert demonstrations and a simulator allowing arbitrary resets can accelerate learning during online RL. We find that training with a suitable choice of an auxiliary start state distribution that may differ from the true start state distribution of the underlying Markov Decision Process can significantly improve sample efficiency. We find that using a notion of safety to inform the choice of this auxiliary distribution significantly accelerates learning. By using episode length information as a way to operationalize this notion, we demonstrate state-of-the-art sample efficiency on a sparse-reward hard-exploration environment.

Jiayu Chen, Le Xu, Aravind Venugopal et al.

Offline reinforcement learning (RL) offers a powerful paradigm for data-driven control. Compared to model-free approaches, offline model-based RL (MBRL) explicitly learns a world model from a static dataset and uses it as a surrogate simulator, improving data efficiency and enabling potential generalization beyond the dataset support. However, most existing offline MBRL methods follow a two-stage training procedure: first learning a world model by maximizing the likelihood of the observed transitions, then optimizing a policy to maximize its expected return under the learned model. This objective mismatch results in a world model that is not necessarily optimized for effective policy learning. Moreover, we observe that policies learned via offline MBRL often lack robustness during deployment, and small adversarial noise in the environment can lead to significant performance degradation. To address these, we propose a framework that dynamically adapts the world model alongside the policy under a unified learning objective aimed at improving robustness. At the core of our method is a maximin optimization problem, which we solve by innovatively utilizing Stackelberg learning dynamics. We provide theoretical analysis to support our design and introduce computationally efficient implementations. We benchmark our algorithm on twelve noisy D4RL MuJoCo tasks and three stochastic Tokamak Control tasks, demonstrating its state-of-the-art performance.

Wen-Tse Chen, Yuxuan Li, Shiyu Huang et al.

Multi-agent credit assignment is a fundamental challenge for cooperative multi-agent reinforcement learning (MARL), where a team of agents learn from shared reward signals. The Individual-Global-Max (IGM) condition is a widely used principle for multi-agent credit assignment, requiring that the joint action determined by individual Q-functions maximizes the global Q-value. Meanwhile, the principle of maximum entropy has been leveraged to enhance exploration in MARL. However, we identify a critical limitation in existing maximum entropy MARL methods: a misalignment arises between local policies and the joint policy that maximizes the global Q-value, leading to violations of the IGM condition. To address this misalignment, we propose an order-preserving transformation. Building on it, we introduce ME-IGM, a novel maximum entropy MARL algorithm compatible with any credit assignment mechanism that satisfies the IGM condition while enjoying the benefits of maximum entropy exploration. We empirically evaluate two variants of ME-IGM: ME-QMIX and ME-QPLEX, in non-monotonic matrix games, and demonstrate their state-of-the-art performance across 17 scenarios in SMAC-v2 and Overcooked.

Arun Balajee Vasudevan, Neehar Peri, Jeff Schneider et al.

Motion planning is crucial for safe navigation in complex urban environments. Historically, motion planners (MPs) have been evaluated with procedurally-generated simulators like CARLA. However, such synthetic benchmarks do not capture real-world multi-agent interactions. nuPlan, a recently released MP benchmark, addresses this limitation by augmenting real-world driving logs with closed-loop simulation logic, effectively turning the fixed dataset into a reactive simulator. We analyze the characteristics of nuPlan's recorded logs and find that each city has its own unique driving behaviors, suggesting that robust planners must adapt to different environments. We learn to model such unique behaviors with BehaviorNet, a graph convolutional neural network (GCNN) that predicts reactive agent behaviors using features derived from recently-observed agent histories; intuitively, some aggressive agents may tailgate lead vehicles, while others may not. To model such phenomena, BehaviorNet predicts the parameters of an agent's motion controller rather than directly predicting its spacetime trajectory (as most forecasters do). Finally, we present AdaptiveDriver, a model-predictive control (MPC) based planner that unrolls different world models conditioned on BehaviorNet's predictions. Our extensive experiments demonstrate that AdaptiveDriver achieves state-of-the-art results on the nuPlan closed-loop planning benchmark, improving over prior work by 2% on Test-14 Hard R-CLS, and generalizes even when evaluated on never-before-seen cities.

Adam Villaflor, Brian Yang, Huangyuan Su et al.

Significant progress has been made in training multimodal trajectory forecasting models for autonomous driving. However, effectively integrating these models with downstream planners and model-based control approaches is still an open problem. Although these models have conventionally been evaluated for open-loop prediction, we show that they can be used to parameterize autoregressive closed-loop models without retraining. We consider recent trajectory prediction approaches which leverage learned anchor embeddings to predict multiple trajectories, finding that these anchor embeddings can parameterize discrete and distinct modes representing high-level driving behaviors. We propose to perform fully reactive closed-loop planning over these discrete latent modes, allowing us to tractably model the causal interactions between agents at each step. We validate our approach on a suite of more dynamic merging scenarios, finding that our approach avoids the $\textit{frozen robot problem}$ which is pervasive in conventional planners. Our approach also outperforms the previous state-of-the-art in CARLA on challenging dense traffic scenarios when evaluated at realistic speeds.

Yeeho Song, Jeff Schneider

Achieving robust performance is crucial when applying deep reinforcement learning (RL) in safety critical systems. Some of the state of the art approaches try to address the problem with adversarial agents, but these agents often require expert supervision to fine tune and prevent the adversary from becoming too challenging to the trainee agent. While other approaches involve automatically adjusting environment setups during training, they have been limited to simple environments where low-dimensional encodings can be used. Inspired by these approaches, we propose genetic curriculum, an algorithm that automatically identifies scenarios in which the agent currently fails and generates an associated curriculum to help the agent learn to solve the scenarios and acquire more robust behaviors. As a non-parametric optimizer, our approach uses a raw, non-fixed encoding of scenarios, reducing the need for expert supervision and allowing our algorithm to adapt to the changing performance of the agent. Our empirical studies show improvement in robustness over the existing state of the art algorithms, providing training curricula that result in agents being 2 - 8x times less likely to fail without sacrificing cumulative reward. We include an ablation study and share insights on why our algorithm outperforms prior approaches.

David Guttendorf, D. W. Wilson Hamilton, Anne Harris Heckman et al.

A robotic system of multiple unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UAVs) has the potential for advancing autonomous object geolocation performance. Much research has focused on algorithmic improvements on individual components, such as navigation, motion planning, and perception. In this paper, we present a UGV-UAV object detection and geolocation system, which performs perception, navigation, and planning autonomously in real scale in unstructured environment. We designed novel sensor pods equipped with multispectral (visible, near-infrared, thermal), high resolution (181.6 Mega Pixels), stereo (near-infrared pair), wide field of view (192 degree HFOV) array. We developed a novel on-board software-hardware architecture to process the high volume sensor data in real-time, and we built a custom AI subsystem composed of detection, tracking, navigation, and planning for autonomous objects geolocation in real-time. This research is the first real scale demonstration of such high speed data processing capability. Our novel modular sensor pod can boost relevant computer vision and machine learning research. Our novel hardware-software architecture is a solid foundation for system-level and component-level research. Our system is validated through data-driven offline tests as well as a series of field tests in unstructured environments. We present quantitative results as well as discussions on key robotic system level challenges which manifest when we build and test the system. This system is the first step toward a UGV-UAV cooperative reconnaissance system in the future.

Benjamin Freed, Aditya Kapoor, Ian Abraham et al.

One of the preeminent obstacles to scaling multi-agent reinforcement learning to large numbers of agents is assigning credit to individual agents' actions. In this paper, we address this credit assignment problem with an approach that we call \textit{partial reward decoupling} (PRD), which attempts to decompose large cooperative multi-agent RL problems into decoupled subproblems involving subsets of agents, thereby simplifying credit assignment. We empirically demonstrate that decomposing the RL problem using PRD in an actor-critic algorithm results in lower variance policy gradient estimates, which improves data efficiency, learning stability, and asymptotic performance across a wide array of multi-agent RL tasks, compared to various other actor-critic approaches. Additionally, we relate our approach to counterfactual multi-agent policy gradient (COMA), a state-of-the-art MARL algorithm, and empirically show that our approach outperforms COMA by making better use of information in agents' reward streams, and by enabling recent advances in advantage estimation to be used.

Viraj Mehta, Biswajit Paria, Jeff Schneider et al.

In many practical applications of RL, it is expensive to observe state transitions from the environment. For example, in the problem of plasma control for nuclear fusion, computing the next state for a given state-action pair requires querying an expensive transition function which can lead to many hours of computer simulation or dollars of scientific research. Such expensive data collection prohibits application of standard RL algorithms which usually require a large number of observations to learn. In this work, we address the problem of efficiently learning a policy while making a minimal number of state-action queries to the transition function. In particular, we leverage ideas from Bayesian optimal experimental design to guide the selection of state-action queries for efficient learning. We propose an acquisition function that quantifies how much information a state-action pair would provide about the optimal solution to a Markov decision process. At each iteration, our algorithm maximizes this acquisition function, to choose the most informative state-action pair to be queried, thus yielding a data-efficient RL approach. We experiment with a variety of simulated continuous control problems and show that our approach learns an optimal policy with up to $5$ -- $1,000\times$ less data than model-based RL baselines and $10^3$ -- $10^5\times$ less data than model-free RL baselines. We also provide several ablated comparisons which point to substantial improvements arising from the principled method of obtaining data.

Divam Gupta, Wei Pu, Trenton Tabor et al.

Accurate layout estimation is crucial for planning and navigation in robotics applications, such as self-driving. In this paper, we introduce the Stereo Bird's Eye ViewNetwork (SBEVNet), a novel supervised end-to-end framework for estimation of bird's eye view layout from a pair of stereo images. Although our network reuses some of the building blocks from the state-of-the-art deep learning networks for disparity estimation, we show that explicit depth estimation is neither sufficient nor necessary. Instead, the learning of a good internal bird's eye view feature representation is effective for layout estimation. Specifically, we first generate a disparity feature volume using the features of the stereo images and then project it to the bird's eye view coordinates. This gives us coarse-grained information about the scene structure. We also apply inverse perspective mapping (IPM) to map the input images and their features to the bird's eye view. This gives us fine-grained texture information. Concatenating IPM features with the projected feature volume creates a rich bird's eye view representation which is useful for spatial reasoning. We use this representation to estimate the BEV semantic map. Additionally, we show that using the IPM features as a supervisory signal for stereo features can give an improvement in performance. We demonstrate our approach on two datasets:the KITTI dataset and a synthetically generated dataset from the CARLA simulator. For both of these datasets, we establish state-of-the-art performance compared to baseline techniques.

Tanmay Agarwal, Hitesh Arora, Jeff Schneider

Traditional autonomous vehicle pipelines that follow a modular approach have been very successful in the past both in academia and industry, which has led to autonomy deployed on road. Though this approach provides ease of interpretation, its generalizability to unseen environments is limited and hand-engineering of numerous parameters is required, especially in the prediction and planning systems. Recently, deep reinforcement learning has been shown to learn complex strategic games and perform challenging robotic tasks, which provides an appealing framework for learning to drive. In this work, we propose a deep reinforcement learning framework to learn optimal control policy using waypoints and low-dimensional visual representations, also known as affordances. We demonstrate that our agents when trained from scratch learn the tasks of lane-following, driving around inter-sections as well as stopping in front of other actors or traffic lights even in the dense traffic setting. We note that our method achieves comparable or better performance than the baseline methods on the original and NoCrash benchmarks on the CARLA simulator.

Youngseog Chung, Willie Neiswanger, Ian Char et al.

Among the many ways of quantifying uncertainty in a regression setting, specifying the full quantile function is attractive, as quantiles are amenable to interpretation and evaluation. A model that predicts the true conditional quantiles for each input, at all quantile levels, presents a correct and efficient representation of the underlying uncertainty. To achieve this, many current quantile-based methods focus on optimizing the so-called pinball loss. However, this loss restricts the scope of applicable regression models, limits the ability to target many desirable properties (e.g. calibration, sharpness, centered intervals), and may produce poor conditional quantiles. In this work, we develop new quantile methods that address these shortcomings. In particular, we propose methods that can apply to any class of regression model, allow for selecting a trade-off between calibration and sharpness, optimize for calibration of centered intervals, and produce more accurate conditional quantiles. We provide a thorough experimental evaluation of our methods, which includes a high dimensional uncertainty quantification task in nuclear fusion.

Ramina Ghods, William J. Durkin, Jeff Schneider

The active search for objects of interest in an unknown environment has many robotics applications including search and rescue, detecting gas leaks or locating animal poachers. Existing algorithms often prioritize the location accuracy of objects of interest while other practical issues such as the reliability of object detection as a function of distance and lines of sight remain largely ignored. Additionally, in many active search scenarios, communication infrastructure may be unreliable or unestablished, making centralized control of multiple agents impractical. We present an algorithm called Noise-Aware Thompson Sampling (NATS) that addresses these issues for multiple ground-based robots performing active search considering two sources of sensory information from monocular optical imagery and depth maps. By utilizing Thompson Sampling, NATS allows for decentralized coordination among multiple agents. NATS also considers object detection uncertainty from depth as well as environmental occlusions and operates while remaining agnostic of the number of objects of interest. Using simulation results, we show that NATS significantly outperforms existing methods such as information-greedy policies or exhaustive search. We demonstrate the real-world viability of NATS using a pseudo-realistic environment created in the Unreal Engine 4 game development platform with the AirSim plugin.