Shlomo Zilberstein

Hyejun Jeong, Amir Houmansadr, Shlomo Zilberstein et al.

Modern AI agents increasingly combine conversational interaction with autonomous task execution, such as coding and web research, raising a natural question: What happens when an agent engaged in long-horizon tasks is exposed to user persuasion? Yet studying this possibility is challenging because long-running agent behavior is noisy and costly to reproduce, and it remains unclear which unique challenges emerge only in extended task execution. We study how belief-level intervention can influence downstream task behavior, a phenomenon we name persuasion propagation. We introduce a behavior-centered evaluation framework that distinguishes between persuasion applied during or prior to task execution. Across web research and coding tasks, we find that on-the-fly persuasion induces weak and inconsistent behavioral effects. In contrast, when the belief state is explicitly specified at task time, belief-prefilled agents conduct on average 26.9% fewer searches and visit 16.9% fewer unique sources than neutral-prefilled agents. These results suggest that persuasion, even in prior interaction, can affect the agent's behavior, motivating behavior-level evaluation in agentic systems.

Samer B. Nashed, Saaduddin Mahmud, Claudia V. Goldman et al.

We introduce a novel framework for causal explanations of stochastic, sequential decision-making systems built on the well-studied structural causal model paradigm for causal reasoning. This single framework can identify multiple, semantically distinct explanations for agent actions -- something not previously possible. In this paper, we establish exact methods and several approximation techniques for causal inference on Markov decision processes using this framework, followed by results on the applicability of the exact methods and some run time bounds. We discuss several scenarios that illustrate the framework's flexibility and the results of experiments with human subjects that confirm the benefits of this approach.

Abhinav Bhatia, Philip S. Thomas, Shlomo Zilberstein

Model-based reinforcement learning promises to learn an optimal policy from fewer interactions with the environment compared to model-free reinforcement learning by learning an intermediate model of the environment in order to predict future interactions. When predicting a sequence of interactions, the rollout length, which limits the prediction horizon, is a critical hyperparameter as accuracy of the predictions diminishes in the regions that are further away from real experience. As a result, with a longer rollout length, an overall worse policy is learned in the long run. Thus, the hyperparameter provides a trade-off between quality and efficiency. In this work, we frame the problem of tuning the rollout length as a meta-level sequential decision-making problem that optimizes the final policy learned by model-based reinforcement learning given a fixed budget of environment interactions by adapting the hyperparameter dynamically based on feedback from the learning process, such as accuracy of the model and the remaining budget of interactions. We use model-free deep reinforcement learning to solve the meta-level decision problem and demonstrate that our approach outperforms common heuristic baselines on two well-known reinforcement learning environments.

Mason Nakamura, Abhinav Kumar, Saswat Das et al.

Multi-agent systems, where LLM agents communicate through free-form language, enable sophisticated coordination for solving complex cooperative tasks. This surfaces a unique safety problem when individual agents form a coalition and \emph{collude} to pursue secondary goals and degrade the joint objective. In this paper, we present Colosseum, a framework for auditing LLM agents' collusive behavior in multi-agent settings. We ground how agents cooperate through a Distributed Constraint Optimization Problem (DCOP) and measure collusion via regret relative to the cooperative optimum. Colosseum tests each LLM for collusion under different objectives, persuasion tactics, and network topologies. Through our audit, we show that most out-of-the-box models exhibited a propensity to collude when a secret communication channel was artificially formed. Furthermore, we discover ``collusion on paper'' when agents plan to collude in text but would often pick non-collusive actions, thus providing little effect on the joint task. Colosseum provides a new way to study collusion by measuring communications and actions in rich yet verifiable environments.

Emily Pruc, Shlomo Zilberstein, Joydeep Biswas

Both pedestrian and robot comfort are of the highest priority whenever a robot is placed in an environment containing human beings. In the case of pedestrian-unaware mobile robots this desire for safety leads to the freezing robot problem, where a robot confronted with a large dynamic group of obstacles (such as a crowd of pedestrians) would determine all forward navigation unsafe causing the robot to stop in place. In order to navigate in a socially compliant manner while avoiding the freezing robot problem we are interested in understanding the flow of pedestrians in crowded scenarios. By treating the pedestrians in the crowd as particles moved along by the crowd itself we can model the system as a time dependent flow field. From this flow field we can extract different flow segments that reflect the motion patterns emerging from the crowd. These motion patterns can then be accounted for during the control and navigation of a mobile robot allowing it to move safely within the flow of the crowd to reach a desired location within or beyond the flow. We combine flow-field extraction with a discrete heuristic search to create Flow-Informed path planning (FIPP). We provide empirical results showing that when compared against a trajectory-rollout local path planner, a robot using FIPP was able not only to reach its goal more quickly but also was shown to be more socially compliant than a robot using traditional techniques both in simulation and on real robots.

Abhinav Bhatia, Samer B. Nashed, Shlomo Zilberstein

Meta reinforcement learning (Meta-RL) methods such as RL$^2$ have emerged as promising approaches for learning data-efficient RL algorithms tailored to a given task distribution. However, they show poor asymptotic performance and struggle with out-of-distribution tasks because they rely on sequence models, such as recurrent neural networks or transformers, to process experiences rather than summarize them using general-purpose RL components such as value functions. In contrast, traditional RL algorithms are data-inefficient as they do not use domain knowledge, but do converge to an optimal policy in the limit. We propose RL$^3$, a principled hybrid approach that incorporates action-values, learned per task via traditional RL, in the inputs to Meta-RL. We show that RL$^3$ earns a greater cumulative reward in the long term compared to RL$^2$ while drastically reducing meta-training time and generalizes better to out-of-distribution tasks. Experiments are conducted on Meta-RL benchmarks and custom discrete domains that exhibit a range of short-term, long-term, and complex dependencies.

Saaduddin Mahmud, Mason Nakamura, Shlomo Zilberstein

The advent of large language models (LLMs) has sparked significant interest in using natural language for preference learning. However, existing methods often suffer from high computational burdens, taxing human supervision, and lack of interpretability. To address these issues, we introduce MAPLE, a framework for large language model-guided Bayesian active preference learning. MAPLE leverages LLMs to model the distribution over preference functions, conditioning it on both natural language feedback and conventional preference learning feedback, such as pairwise trajectory rankings. MAPLE also employs active learning to systematically reduce uncertainty in this distribution and incorporates a language-conditioned active query selection mechanism to identify informative and easy-to-answer queries, thus reducing human burden. We evaluate MAPLE's sample efficiency and preference inference quality across two benchmarks, including a real-world vehicle route planning benchmark using OpenStreetMap data. Our results demonstrate that MAPLE accelerates the learning process and effectively improves humans' ability to answer queries.

Saaduddin Mahmud, Dorian Benhamou Goldfajn, Shlomo Zilberstein

Distributed Constraint Optimization Problems (DCOPs) offer a powerful framework for multi-agent coordination but often rely on labor-intensive, manual problem construction. To address this, we introduce VL-DCOPs, a framework that takes advantage of large multimodal foundation models (LFMs) to automatically generate constraints from both visual and linguistic instructions. We then introduce a spectrum of agent archetypes for solving VL-DCOPs: from a neuro-symbolic agent that delegates some of the algorithmic decisions to an LFM, to a fully neural agent that depends entirely on an LFM for coordination. We evaluate these agent archetypes using state-of-the-art LLMs (large language models) and VLMs (vision language models) on three novel VL-DCOP tasks and compare their respective advantages and drawbacks. Lastly, we discuss how this work extends to broader frontier challenges in the DCOP literature.

Mason Nakamura, Abhinav Kumar, Saaduddin Mahmud et al.

A multi-agent system (MAS) powered by large language models (LLMs) can automate tedious user tasks such as meeting scheduling that requires inter-agent collaboration. LLMs enable nuanced protocols that account for unstructured private data, user constraints, and preferences. However, this design introduces new risks, including misalignment and attacks by malicious parties that compromise agents or steal user data. In this paper, we propose the Terrarium framework for fine-grained study on safety, privacy, and security in LLM-based MAS. We repurpose the blackboard design, an early approach in multi-agent systems, to create a modular, configurable testbed for multi-agent collaboration. We identify key attack vectors such as misalignment, malicious agents, compromised communication, and data poisoning. We implement three collaborative MAS scenarios with four representative attacks to demonstrate the framework's flexibility. By providing tools to rapidly prototype, evaluate, and iterate on defenses and designs, Terrarium aims to accelerate progress toward trustworthy multi-agent systems.

Saaduddin Mahmud, Eugene Bagdasarian, Shlomo Zilberstein

Agents powered by large language models (LLMs) are increasingly deployed in settings where communication shapes high-stakes decisions, making a principled understanding of strategic communication essential. Prior work largely studies either unverifiable cheap-talk or fully verifiable disclosure, failing to capture realistic domains in which information has probabilistic credibility. We introduce MixTalk, a strategic communication game for LLM-to-LLM interaction that models information credibility. In MixTalk, a sender agent strategically combines verifiable and unverifiable claims to communicate private information, while a receiver agent allocates a limited budget to costly verification and infers the underlying state from prior beliefs, claims, and verification outcomes. We evaluate state-of-the-art LLM agents in large-scale tournaments across three realistic deployment settings, revealing their strengths and limitations in reasoning about information credibility and the explicit behavior that shapes these interactions. Finally, we propose Tournament Oracle Policy Distillation (TOPD), an offline method that distills tournament oracle policy from interaction logs and deploys it in-context at inference time. Our results show that TOPD significantly improves receiver robustness to persuasion.

Saaduddin Mahmud, Mason Nakamura, Kyle H. Wray et al.

Prompt optimization methods have demonstrated significant effectiveness in aligning black-box large language models (LLMs). In parallel, inference scaling strategies such as Best-of-N Sampling and Majority Voting have also proven to enhance alignment and performance by trading off computation. However, existing prompt optimization approaches are inference strategy agnostic; that is, they optimize prompts without regard to the inference strategy employed during deployment. This constitutes a significant methodological gap, as our empirical and theoretical analysis reveals a strong interdependence between these two paradigms. Moreover, we find that user preferences regarding trade-offs among multiple objectives and inference budgets substantially influence the choice of prompt and inference configuration. To address this gap, we introduce a unified novel framework named IAPO (Inference-Aware Prompt Optimization) that jointly optimizes the prompt and inference scale, while being aware of the inference budget and different task objectives. We then develop a fixed-budget training algorithm for IAPO, which we call PSST (Prompt Scaling via Sequential Trimming), and analyze finite-budget guarantees on error probability. Finally, we evaluate the effectiveness of PSST on six different tasks, including multi-objective text generation and reasoning, and demonstrate the critical role of incorporating inference-awareness when aligning black-box LLMs through prompt optimization.

Mason Nakamura, Saaduddin Mahmud, Kyle H. Wray et al.

Aligning LLMs with user preferences is crucial for real-world use but often requires costly fine-tuning or expensive inference, forcing trade-offs between alignment quality and computational cost. Existing inference-time methods typically ignore this balance, focusing solely on the optimized policy's performance. We propose HIA (Heuristic-Guided Inference-time Alignment), a tuning-free, black-box-compatible approach that uses a lightweight prompt optimizer, heuristic reward models, and two-stage filtering to reduce inference calls while preserving alignment quality. On real-world prompt datasets, HelpSteer and ComPRed, HIA outperforms best-of-N sampling, beam search, and greedy search baselines in multi-objective, goal-conditioned tasks under the same inference budget. We also find that HIA is effective under low-inference budgets with as little as one or two response queries, offering a practical solution for scalable, personalized LLM deployment.

Mouad Abrini, Omri Abend, Dina Acklin et al. · cambridge

This volume includes a selection of papers presented at the Workshop on Advancing Artificial Intelligence through Theory of Mind held at AAAI 2025 in Philadelphia US on 3rd March 2025. The purpose of this volume is to provide an open access and curated anthology for the ToM and AI research community.

Sadegh Rabiee, Connor Basich, Kyle Hollins Wray et al.

Robots deployed in the real world over extended periods of time need to reason about unexpected failures, learn to predict them, and to proactively take actions to avoid future failures. Existing approaches for competence-aware planning are either model-based, requiring explicit enumeration of known failure modes, or purely statistical, using state- and location-specific failure statistics to infer competence. We instead propose a structured model-free approach to competence-aware planning by reasoning about plan execution failures due to errors in perception, without requiring a priori enumeration of failure sources or requiring location-specific failure statistics. We introduce competence-aware path planning via introspective perception (CPIP), a Bayesian framework to iteratively learn and exploit task-level competence in novel deployment environments. CPIP factorizes the competence-aware planning problem into two components. First, perception errors are learned in a model-free and location-agnostic setting via introspective perception prior to deployment in novel environments. Second, during actual deployments, the prediction of task-level failures is learned in a context-aware setting. Experiments in a simulation show that the proposed CPIP approach outperforms the frequentist baseline in multiple mobile robot tasks, and is further validated via real robot experiments in an environment with perceptually challenging obstacles and terrain.

Shane Parr, Ishan Khatri, Justin Svegliato et al.

Autonomous systems often operate in environments where the behavior of multiple agents is coordinated by a shared global state. Reliable estimation of the global state is thus critical for successfully operating in a multi-agent setting. We introduce agent-aware state estimation -- a framework for calculating indirect estimations of state given observations of the behavior of other agents in the environment. We also introduce transition-independent agent-aware state estimation -- a tractable class of agent-aware state estimation -- and show that it allows the speed of inference to scale linearly with the number of agents in the environment. As an example, we model traffic light classification in instances of complete loss of direct observation. By taking into account observations of vehicular behavior from multiple directions of traffic, our approach exhibits accuracy higher than that of existing traffic light-only HMM methods on a real-world autonomous vehicle data set under a variety of simulated occlusion scenarios.

Sandhya Saisubramanian, Shlomo Zilberstein

Agents operating in unstructured environments often produce negative side effects (NSE), which are difficult to identify at design time. While the agent can learn to mitigate the side effects from human feedback, such feedback is often expensive and the rate of learning is sensitive to the agent's state representation. We examine how humans can assist an agent, beyond providing feedback, and exploit their broader scope of knowledge to mitigate the impacts of NSE. We formulate this problem as a human-agent team with decoupled objectives. The agent optimizes its assigned task, during which its actions may produce NSE. The human shapes the environment through minor reconfiguration actions so as to mitigate the impacts of the agent's side effects, without affecting the agent's ability to complete its assigned task. We present an algorithm to solve this problem and analyze its theoretical properties. Through experiments with human subjects, we assess the willingness of users to perform minor environment modifications to mitigate the impacts of NSE. Empirical evaluation of our approach shows that the proposed framework can successfully mitigate NSE, without affecting the agent's ability to complete its assigned task.

Sainyam Galhotra, Sandhya Saisubramanian, Shlomo Zilberstein

Fair clustering is the process of grouping similar entities together, while satisfying a mathematically well-defined fairness metric as a constraint. Due to the practical challenges in precise model specification, the prescribed fairness constraints are often incomplete and act as proxies to the intended fairness requirement, leading to biased outcomes when the system is deployed. We examine how to identify the intended fairness constraint for a problem based on limited demonstrations from an expert. Each demonstration is a clustering over a subset of the data. We present an algorithm to identify the fairness metric from demonstrations and generate clusters using existing off-the-shelf clustering techniques, and analyze its theoretical properties. To extend our approach to novel fairness metrics for which clustering algorithms do not currently exist, we present a greedy method for clustering. Additionally, we investigate how to generate interpretable solutions using our approach. Empirical evaluation on three real-world datasets demonstrates the effectiveness of our approach in quickly identifying the underlying fairness and interpretability constraints, which are then used to generate fair and interpretable clusters.

Richard G. Freedman, Steven J. Levine, Brian C. Williams et al.

As robotic teammates become more common in society, people will assess the robots' roles in their interactions along many dimensions. One such dimension is effectiveness: people will ask whether their robotic partners are trustworthy and effective collaborators. This begs a crucial question: how can we quantitatively measure the helpfulness of a robotic partner for a given task at hand? This paper seeks to answer this question with regards to the interactive robot's decision making. We describe a clear, concise, and task-oriented metric applicable to many different planning and execution paradigms. The proposed helpfulness metric is fundamental to assessing the benefit that a partner has on a team for a given task. In this paper, we define helpfulness, illustrate it on concrete examples from a variety of domains, discuss its properties and ramifications for planning interactions with humans, and present preliminary results.

Sandhya Saisubramanian, Shlomo Zilberstein, Ece Kamar

Autonomous agents acting in the real-world often operate based on models that ignore certain aspects of the environment. The incompleteness of any given model -- handcrafted or machine acquired -- is inevitable due to practical limitations of any modeling technique for complex real-world settings. Due to the limited fidelity of its model, an agent's actions may have unexpected, undesirable consequences during execution. Learning to recognize and avoid such negative side effects of an agent's actions is critical to improve the safety and reliability of autonomous systems. Mitigating negative side effects is an emerging research topic that is attracting increased attention due to the rapid growth in the deployment of AI systems and their broad societal impacts. This article provides a comprehensive overview of different forms of negative side effects and the recent research efforts to address them. We identify key characteristics of negative side effects, highlight the challenges in avoiding negative side effects, and discuss recently developed approaches, contrasting their benefits and limitations. The article concludes with a discussion of open questions and suggestions for future research directions.

Connor Basich, Justin Svegliato, Kyle Hollins Wray et al.

Given the complexity of real-world, unstructured domains, it is often impossible or impractical to design models that include every feature needed to handle all possible scenarios that an autonomous system may encounter. For an autonomous system to be reliable in such domains, it should have the ability to improve its competence online. In this paper, we propose a method for improving the competence of a system over the course of its deployment. We specifically focus on a class of semi-autonomous systems known as competence-aware systems that model their own competence -- the optimal extent of autonomy to use in any given situation -- and learn this competence over time from feedback received through interactions with a human authority. Our method exploits such feedback to identify important state features missing from the system's initial model, and incorporates them into its state representation. The result is an agent that better predicts human involvement, leading to improvements in its competence and reliability, and as a result, its overall performance.

Connor Basich, Justin Svegliato, Kyle Hollins Wray et al.

Interest in semi-autonomous systems (SAS) is growing rapidly as a paradigm to deploy autonomous systems in domains that require occasional reliance on humans. This paradigm allows service robots or autonomous vehicles to operate at varying levels of autonomy and offer safety in situations that require human judgment. We propose an introspective model of autonomy that is learned and updated online through experience and dictates the extent to which the agent can act autonomously in any given situation. We define a competence-aware system (CAS) that explicitly models its own proficiency at different levels of autonomy and the available human feedback. A CAS learns to adjust its level of autonomy based on experience to maximize overall efficiency, factoring in the cost of human assistance. We analyze the convergence properties of CAS and provide experimental results for robot delivery and autonomous driving domains that demonstrate the benefits of the approach.

Sandhya Saisubramanian, Sainyam Galhotra, Shlomo Zilberstein

Graph clustering groups entities -- the vertices of a graph -- based on their similarity, typically using a complex distance function over a large number of features. Successful integration of clustering approaches in automated decision-support systems hinges on the interpretability of the resulting clusters. This paper addresses the problem of generating interpretable clusters, given features of interest that signify interpretability to an end-user, by optimizing interpretability in addition to common clustering objectives. We propose a $β$-interpretable clustering algorithm that ensures that at least $β$ fraction of nodes in each cluster share the same feature value. The tunable parameter $β$ is user-specified. We also present a more efficient algorithm for scenarios with $β\!=\!1$ and analyze the theoretical guarantees of the two algorithms. Finally, we empirically demonstrate the benefits of our approaches in generating interpretable clusters using four real-world datasets. The interpretability of the clusters is complemented by generating simple explanations denoting the feature values of the nodes in the clusters, using frequent pattern mining.

Richard G. Freedman, Yi Ren Fung, Roman Ganchin et al.

Many intelligent systems currently interact with others using at least one of fixed communication inputs or preset responses, resulting in rigid interaction experiences and extensive efforts developing a variety of scenarios for the system. Fixed inputs limit the natural behavior of the user in order to effectively communicate, and preset responses prevent the system from adapting to the current situation unless it was specifically implemented. Closed-loop interaction instead focuses on dynamic responses that account for what the user is currently doing based on interpretations of their perceived activity. Agents employing closed-loop interaction can also monitor their interactions to ensure that the user responds as expected. We introduce a closed-loop interactive agent framework that integrates planning and recognition to predict what the user is trying to accomplish and autonomously decide on actions to take in response to these predictions. Based on a recent demonstration of such an assistive interactive agent in a turn-based simulated game, we also discuss new research challenges that are not present in the areas of artificial intelligence planning or recognition alone.

Sandhya Saisubramanian, Shlomo Zilberstein

Reduced models of large Markov decision processes accelerate planning by considering a subset of outcomes for each state-action pair. This reduction in reachable states leads to replanning when the agent encounters states without a precomputed action during plan execution. However, not all states are suitable for replanning. In the worst case, the agent may not be able to reach the goal from the newly encountered state. Agents should be better prepared to handle such risky situations and avoid replanning in risky states. Hence, we consider replanning in states that are unsafe for deliberation as a negative side effect of planning with reduced models. While the negative side effects can be minimized by always using the full model, this defeats the purpose of using reduced models. The challenge is to plan with reduced models, but somehow account for the possibility of encountering risky situations. An agent should thus only replan in states that the user has approved as safe for replanning. To that end, we propose planning using a portfolio of reduced models, a planning paradigm that minimizes the negative side effects of planning using reduced models by alternating between different outcome selection approaches. We empirically demonstrate the effectiveness of our approach on three domains: an electric vehicle charging domain using real-world data from a university campus and two benchmark planning problems.

Sainyam Galhotra, Sandhya Saisubramanian, Shlomo Zilberstein

We introduce a rich model for multi-objective clustering with lexicographic ordering over objectives and a slack. The slack denotes the allowed multiplicative deviation from the optimal objective value of the higher priority objective to facilitate improvement in lower-priority objectives. We then propose an algorithm called Zeus to solve this class of problems, which is characterized by a makeshift function. The makeshift fine tunes the clusters formed by the processed objectives so as to improve the clustering with respect to the unprocessed objectives, given the slack. We present makeshift for solving three different classes of objectives and analyze their solution guarantees. Finally, we empirically demonstrate the effectiveness of our approach on three applications using real-world data.

Sandhya Saisubramanian, Kyle Hollins Wray, Luis Pineda et al.

We present the Goal Uncertain Stochastic Shortest Path (GUSSP) problem -- a general framework to model path planning and decision making in stochastic environments with goal uncertainty. The framework extends the stochastic shortest path (SSP) model to dynamic environments in which it is impossible to determine the exact goal states ahead of plan execution. GUSSPs introduce flexibility in goal specification by allowing a belief over possible goal configurations. The unique observations at potential goals helps the agent identify the true goal during plan execution. The partial observability is restricted to goals, facilitating the reduction to an SSP with a modified state space. We formally define a GUSSP and discuss its theoretical properties. We then propose an admissible heuristic that reduces the planning time using FLARES -- a start-of-the-art probabilistic planner. We also propose a determinization approach for solving this class of problems. Finally, we present empirical results on a search and rescue mobile robot and three other problem domains in simulation.

Siddharth Srivastava, Nishant Desai, Richard Freedman et al.

Integrated task and motion planning has emerged as a challenging problem in sequential decision making, where a robot needs to compute high-level strategy and low-level motion plans for solving complex tasks. While high-level strategies require decision making over longer time-horizons and scales, their feasibility depends on low-level constraints based upon the geometries and continuous dynamics of the environment. The hybrid nature of this problem makes it difficult to scale; most existing approaches focus on deterministic, fully observable scenarios. We present a new approach where the high-level decision problem occurs in a stochastic setting and can be modeled as a Markov decision process. In contrast to prior efforts, we show that complete MDP policies, or contingent behaviors, can be computed effectively in an anytime fashion. Our algorithm continuously improves the quality of the solution and is guaranteed to be probabilistically complete. We evaluate the performance of our approach on a challenging, realistic test problem: autonomous aircraft inspection. Our results show that we can effectively compute consistent task and motion policies for the most likely execution-time outcomes using only a fraction of the computation required to develop the complete task and motion policy.

Luis Pineda, Shlomo Zilberstein

The stochastic shortest path problem (SSP) is a highly expressive model for probabilistic planning. The computational hardness of SSPs has sparked interest in determinization-based planners that can quickly solve large problems. However, existing methods employ a simplistic approach to determinization. In particular, they ignore the possibility of tailoring the determinization to the specific characteristics of the target domain. In this work we examine this question, by showing that learning a good determinization for a planning domain can be done efficiently and can improve performance. Moreover, we show how to directly incorporate probabilistic reasoning into the planning problem when a good determinization is not sufficient by itself. Based on these insights, we introduce a planner, FF-LAO*, that outperforms state-of-the-art probabilistic planners on several well-known competition benchmarks.

Xiaojian Wu, Akshat Kumar, Daniel Sheldon et al.

Stochastic network design is a general framework for optimizing network connectivity. It has several applications in computational sustainability including spatial conservation planning, pre-disaster network preparation, and river network optimization. A common assumption in previous work has been made that network parameters (e.g., probability of species colonization) are precisely known, which is unrealistic in real- world settings. We therefore address the robust river network design problem where the goal is to optimize river connectivity for fish movement by removing barriers. We assume that fish passability probabilities are known only imprecisely, but are within some interval bounds. We then develop a planning approach that computes the policies with either high robust ratio or low regret. Empirically, our approach scales well to large river networks. We also provide insights into the solutions generated by our robust approach, which has significantly higher robust ratio than the baseline solution with mean parameter estimates.

Marek Petrik, Shlomo Zilberstein

Multiagent planning and coordination problems are common and known to be computationally hard. We show that a wide range of two-agent problems can be formulated as bilinear programs. We present a successive approximation algorithm that significantly outperforms the coverage set algorithm, which is the state-of-the-art method for this class of multiagent problems. Because the algorithm is formulated for bilinear programs, it is more general and simpler to implement. The new algorithm can be terminated at any time and-unlike the coverage set algorithm-it facilitates the derivation of a useful online performance bound. It is also much more efficient, on average reducing the computation time of the optimal solution by about four orders of magnitude. Finally, we introduce an automatic dimensionality reduction method that improves the effectiveness of the algorithm, extending its applicability to new domains and providing a new way to analyze a subclass of bilinear programs.

Daniel S. Bernstein, Christopher Amato, Eric A. Hansen et al.

Coordination of distributed agents is required for problems arising in many areas, including multi-robot systems, networking and e-commerce. As a formal framework for such problems, we use the decentralized partially observable Markov decision process (DEC-POMDP). Though much work has been done on optimal dynamic programming algorithms for the single-agent version of the problem, optimal algorithms for the multiagent case have been elusive. The main contribution of this paper is an optimal policy iteration algorithm for solving DEC-POMDPs. The algorithm uses stochastic finite-state controllers to represent policies. The solution can include a correlation device, which allows agents to correlate their actions without communicating. This approach alternates between expanding the controller and performing value-preserving transformations, which modify the controller without sacrificing value. We present two efficient value-preserving transformations: one can reduce the size of the controller and the other can improve its value while keeping the size fixed. Empirical results demonstrate the usefulness of value-preserving transformations in increasing value while keeping controller size to a minimum. To broaden the applicability of the approach, we also present a heuristic version of the policy iteration algorithm, which sacrifices convergence to optimality. This algorithm further reduces the size of the controllers at each step by assuming that probability distributions over the other agents actions are known. While this assumption may not hold in general, it helps produce higher quality solutions in our test problems.

Daniel S Bernstein, Shlomo Zilberstein, Neil Immerman

Planning for distributed agents with partial state information is considered from a decision- theoretic perspective. We describe generalizations of both the MDP and POMDP models that allow for decentralized control. For even a small number of agents, the finite-horizon problems corresponding to both of our models are complete for nondeterministic exponential time. These complexity results illustrate a fundamental difference between centralized and decentralized control of Markov processes. In contrast to the MDP and POMDP problems, the problems we consider provably do not admit polynomial-time algorithms and most likely require doubly exponential time to solve in the worst case. We have thus provided mathematical evidence corresponding to the intuition that decentralized planning problems cannot easily be reduced to centralized problems and solved exactly using established techniques.

Zhengzhu Feng, Eric A. Hansen, Shlomo Zilberstein

Symbolic representations have been used successfully in off-line planning algorithms for Markov decision processes. We show that they can also improve the performance of on-line planners. In addition to reducing computation time, symbolic generalization can reduce the amount of costly real-world interactions required for convergence. We introduce Symbolic Real-Time Dynamic Programming (or sRTDP), an extension of RTDP. After each step of on-line interaction with an environment, sRTDP uses symbolic model-checking techniques to generalizes its experience by updating a group of states rather than a single state. We examine two heuristic approaches to dynamic grouping of states and show that they accelerate the planning process significantly in terms of both CPU time and the number of steps of interaction with the environment.

Zhengzhu Feng, Shlomo Zilberstein

We present a major improvement to the incremental pruning algorithm for solving partially observable Markov decision processes. Our technique targets the cross-sum step of the dynamic programming (DP) update, a key source of complexity in POMDP algorithms. Instead of reasoning about the whole belief space when pruning the cross-sums, our algorithm divides the belief space into smaller regions and performs independent pruning in each region. We evaluate the benefits of the new technique both analytically and experimentally, and show that it produces very significant performance gains. The results contribute to the scalability of POMDP algorithms to domains that cannot be handled by the best existing techniques.

Daniel Szer, Francois Charpillet, Shlomo Zilberstein

We present multi-agent A* (MAA*), the first complete and optimal heuristic search algorithm for solving decentralized partially-observable Markov decision problems (DEC-POMDPs) with finite horizon. The algorithm is suitable for computing optimal plans for a cooperative group of agents that operate in a stochastic environment such as multirobot coordination, network traffic control, `or distributed resource allocation. Solving such problems efiectively is a major challenge in the area of planning under uncertainty. Our solution is based on a synthesis of classical heuristic search and decentralized control theory. Experimental results show that MAA* has significant advantages. We introduce an anytime variant of MAA* and conclude with a discussion of promising extensions such as an approach to solving infinite horizon problems.

Sven Seuken, Shlomo Zilberstein

Memory-Bounded Dynamic Programming (MBDP) has proved extremely effective in solving decentralized POMDPs with large horizons. We generalize the algorithm and improve its scalability by reducing the complexity with respect to the number of observations from exponential to polynomial. We derive error bounds on solution quality with respect to this new approximation and analyze the convergence behavior. To evaluate the effectiveness of the improvements, we introduce a new, larger benchmark problem. Experimental results show that despite the high complexity of decentralized POMDPs, scalable solution techniques such as MBDP perform surprisingly well.

Christopher Amato, Daniel S Bernstein, Shlomo Zilberstein

We present a memory-bounded optimization approach for solving infinite-horizon decentralized POMDPs. Policies for each agent are represented by stochastic finite state controllers. We formulate the problem of optimizing these policies as a nonlinear program, leveraging powerful existing nonlinear optimization techniques for solving the problem. While existing solvers only guarantee locally optimal solutions, we show that our formulation produces higher quality controllers than the state-of-the-art approach. We also incorporate a shared source of randomness in the form of a correlation device to further increase solution quality with only a limited increase in space and time. Our experimental results show that nonlinear optimization can be used to provide high quality, concise solutions to decentralized decision problems under uncertainty.

Feng Wu, Shlomo Zilberstein, Xiaoping Chen

We present decentralized rollout sampling policy iteration (DecRSPI) - a new algorithm for multi-agent decision problems formalized as DEC-POMDPs. DecRSPI is designed to improve scalability and tackle problems that lack an explicit model. The algorithm uses Monte- Carlo methods to generate a sample of reachable belief states. Then it computes a joint policy for each belief state based on the rollout estimations. A new policy representation allows us to represent solutions compactly. The key benefits of the algorithm are its linear time complexity over the number of agents, its bounded memory usage and good solution quality. It can solve larger problems that are intractable for existing planning algorithms. Experimental results confirm the effectiveness and scalability of the approach.

Akshat Kumar, Shlomo Zilberstein

Decentralized POMDPs provide an expressive framework for multi-agent sequential decision making. While fnite-horizon DECPOMDPs have enjoyed signifcant success, progress remains slow for the infnite-horizon case mainly due to the inherent complexity of optimizing stochastic controllers representing agent policies. We present a promising new class of algorithms for the infnite-horizon case, which recasts the optimization problem as inference in a mixture of DBNs. An attractive feature of this approach is the straightforward adoption of existing inference techniques in DBNs for solving DEC-POMDPs and supporting richer representations such as factored or continuous states and actions. We also derive the Expectation Maximization (EM) algorithm to optimize the joint policy represented as DBNs. Experiments on benchmark domains show that EM compares favorably against the state-of-the-art solvers.

Akshat Kumar, Shlomo Zilberstein

Computing maximum a posteriori (MAP) estimation in graphical models is an important inference problem with many applications. We present message-passing algorithms for quadratic programming (QP) formulations of MAP estimation for pairwise Markov random fields. In particular, we use the concave-convex procedure (CCCP) to obtain a locally optimal algorithm for the non-convex QP formulation. A similar technique is used to derive a globally convergent algorithm for the convex QP relaxation of MAP. We also show that a recently developed expectation-maximization (EM) algorithm for the QP formulation of MAP can be derived from the CCCP perspective. Experiments on synthetic and real-world problems confirm that our new approach is competitive with max-product and its variations. Compared with CPLEX, we achieve more than an order-of-magnitude speedup in solving optimally the convex QP relaxation.