A large amount of work has been done in Multi-Agent Systems (MAS) for modeling and solving problems with multiple interacting agents. However, most LLMs are pretrained independently and not specifically optimized for coordination. Existing LLM fine-tuning frameworks rely on individual rewards, which require complex reward designs for each agent to encourage collaboration. To address these challenges, we model LLM collaboration as a cooperative Multi-Agent Reinforcement Learning (MARL) problem. We develop a multi-agent, multi-turn algorithm, Multi-Agent Group Relative Policy Optimization (MAGRPO), to solve it, building on current RL approaches for LLMs as well as MARL techniques. Our experiments on LLM writing and coding collaboration demonstrate that fine-tuning MAS with MAGRPO enables agents to generate high-quality responses efficiently through effective cooperation. Our approach opens the door to using MARL methods for LLM collaboration and highlights the associated challenges.

Ensuring safety in reinforcement learning (RL) is critical for deploying agents in real-world applications. During training, current safe RL approaches often rely on indicator cost functions that provide sparse feedback, resulting in two key limitations: (i) poor sample efficiency due to the lack of safety information in neighboring states, and (ii) dependence on cost-value functions, leading to brittle convergence and suboptimal performance. After training, safety is guaranteed via formal verification (FV) methods for deep neural networks, whose computational complexity hinders their application during training. We address the limitations of using cost functions via verification by proposing a safe RL method based on a violation value—the risk associated with policy decisions in a portion of the state space. Our approach verifies safety properties (i.e., state-action pairs) that may lead to unsafe behavior, and quantifies the size of the state space where properties are violated. This violation value is then used to penalize the agent during training to encourage safer policy behavior. Given the NP-hard nature of FV, we propose an efficient, sample-based approximation with probabilistic guarantees to compute the violation value. Extensive experiments on standard benchmarks and real-world robotic navigation tasks show that violation-augmented approaches significantly improve safety by reducing the number of unsafe states encountered while achieving superior performance compared to existing methods.

Recent works have demonstrated the vulnerability of Deep Reinforcement Learning (DRL) algorithms against training-time, backdoor poisoning attacks. The objectives of these attacks are twofold: induce pre-determined, adversarial behavior in the agent upon observing a fixed trigger during deployment while allowing the agent to solve its intended task during training. Prior attacks assume arbitrary control over the agent's rewards, inducing values far outside the environment's natural constraints. This results in brittle attacks that fail once the proper reward constraints are enforced. Thus, in this work we propose a new class of backdoor attacks against DRL which are the first to achieve state of the art performance under strict reward constraints. These ``inception'' attacks manipulate the agent's training data -- inserting the trigger into prior observations and replacing high return actions with those of the targeted adversarial behavior. We formally define these attacks and prove they achieve both adversarial objectives against arbitrary Markov Decision Processes (MDP). Using this framework we devise an online inception attack which achieves an 100% attack success rate on multiple environments under constrained rewards while minimally impacting the agent's task performance.

Metric occupancy maps are widely used in autonomous robot navigation systems. However, when a robot is deployed in an unseen environment, building an accurate metric map is time-consuming. Can an autonomous robot directly navigate in previously unseen environments using coarse maps? In this work, we propose the Coarse Map Navigator (CMN), a navigation framework that can perform robot navigation in unseen environments using different coarse maps. To do so, CMN addresses two challenges: (1) novel and realistic visual observations; (2) error and misalignment on coarse maps. To tackle novel visual observations in unseen environments, CMN learns a deep perception model that maps the visual input from various pixel spaces to the local occupancy grid space. To tackle the error and misalignment on coarse maps, CMN extends the Bayesian filter and maintains a belief directly on coarse maps using the predicted local occupancy grids as observations. Using the latest belief, CMN extracts a global heuristic vector that guides the planner to find a local navigation action. Empirical results demonstrate that CMN achieves high navigation success rates in unseen environments, significantly outperforming baselines, and is robust to different coarse maps.

Reinforcement learning (RL) is an actively growing field that is seeing increased usage in real-world, safety-critical applications -- making it paramount to ensure the robustness of RL algorithms against adversarial attacks. In this work we explore a particularly stealthy form of training-time attacks against RL -- backdoor poisoning. Here the adversary intercepts the training of an RL agent with the goal of reliably inducing a particular action when the agent observes a pre-determined trigger at inference time. We uncover theoretical limitations of prior work by proving their inability to generalize across domains and MDPs. Motivated by this, we formulate a novel poisoning attack framework which interlinks the adversary's objectives with those of finding an optimal policy -- guaranteeing attack success in the limit. Using insights from our theoretical analysis we develop "SleeperNets" as a universal backdoor attack which exploits a newly proposed threat model and leverages dynamic reward poisoning techniques. We evaluate our attack in 6 environments spanning multiple domains and demonstrate significant improvements in attack success over existing methods, while preserving benign episodic return.

Off-policy learning from multistep returns is crucial for sample-efficient reinforcement learning, but counteracting off-policy bias without exacerbating variance is challenging. Classically, off-policy bias is corrected in a per-decision manner: past temporal-difference errors are re-weighted by the instantaneous Importance Sampling (IS) ratio after each action via eligibility traces. Many off-policy algorithms rely on this mechanism, along with differing protocols for cutting the IS ratios (traces) to combat the variance of the IS estimator. Unfortunately, once a trace has been cut, the effect cannot be easily reversed. This has led to the development of credit-assignment strategies that account for multiple past experiences at a time. These trajectory-aware methods have not been extensively analyzed, and their theoretical justification remains uncertain. In this paper, we propose a multistep operator that unifies per-decision and trajectory-aware methods. We prove convergence conditions for our operator in the tabular setting, establishing the first guarantees for several existing methods as well as many new ones. Finally, we introduce Recency-Bounded Importance Sampling (RBIS), which leverages trajectory awareness to perform robustly across $\lambda$-values in an off-policy control task.

Robot learning is often difficult due to the expense of gathering data. The need for large amounts of data can, and should, be tackled with effective algorithms and leveraging expert information on robot dynamics. Bayesian reinforcement learning (BRL), thanks to its sample efficiency and ability to exploit prior knowledge, is uniquely positioned as such a solution method. Unfortunately, the application of BRL has been limited due to the difficulties of representing expert knowledge as well as solving the subsequent inference problem. This paper advances BRL for robotics by proposing a specialized framework for physical systems. In particular, we capture this knowledge in a factored representation, then demonstrate the posterior factorizes in a similar shape, and ultimately formalize the model in a Bayesian framework. We then introduce a sample-based online solution method, based on Monte-Carlo tree search and particle filtering, specialized to solve the resulting model. This approach can, for example, utilize typical low-level robot simulators and handle uncertainty over unknown dynamics of the environment. We empirically demonstrate its efficiency by performing on-robot learning in two human-robot interaction tasks with uncertainty about human behavior, achieving near-optimal performance after only a handful of real-world episodes. A video of learned policies is at https://youtu.be/H9xp60ngOes.

Centralized Training for Decentralized Execution, where agents are trained offline in a centralized fashion and execute online in a decentralized manner, has become a popular approach in Multi-Agent Reinforcement Learning (MARL). In particular, it has become popular to develop actor-critic methods that train decentralized actors with a centralized critic where the centralized critic is allowed access global information of the entire system, including the true system state. Such centralized critics are possible given offline information and are not used for online execution. While these methods perform well in a number of domains and have become a de facto standard in MARL, using a centralized critic in this context has yet to be sufficiently analyzed theoretically or empirically. In this paper, we therefore formally analyze centralized and decentralized critic approaches, and analyze the effect of using state-based critics in partially observable environments. We derive theories contrary to the common intuition: critic centralization is not strictly beneficial, and using state values can be harmful. We further prove that, in particular, state-based critics can introduce unexpected bias and variance compared to history-based critics. Finally, we demonstrate how the theory applies in practice by comparing different forms of critics on a wide range of common multi-agent benchmarks. The experiments show practical issues such as the difficulty of representation learning with partial observability, which highlights why the theoretical problems are often overlooked in the literature.

Centralized Training for Decentralized Execution, where training is done in a centralized offline fashion, has become a popular solution paradigm in Multi-Agent Reinforcement Learning. Many such methods take the form of actor-critic with state-based critics, since centralized training allows access to the true system state, which can be useful during training despite not being available at execution time. State-based critics have become a common empirical choice, albeit one which has had limited theoretical justification or analysis. In this paper, we show that state-based critics can introduce bias in the policy gradient estimates, potentially undermining the asymptotic guarantees of the algorithm. We also show that, even if the state-based critics do not introduce any bias, they can still result in a larger gradient variance, contrary to the common intuition. Finally, we show the effects of the theories in practice by comparing different forms of centralized critics on a wide range of common benchmarks, and detail how various environmental properties are related to the effectiveness of different types of critics.

Learning safe solutions is an important but challenging problem in multi-agent reinforcement learning (MARL). Shielded reinforcement learning is one approach for preventing agents from choosing unsafe actions. Current shielded reinforcement learning methods for MARL make strong assumptions about communication and full observability. In this work, we extend the formalization of the shielded reinforcement learning problem to a decentralized multi-agent setting. We then present an algorithm for decomposition of a centralized shield, allowing shields to be used in such decentralized, communication-free environments. Our results show that agents equipped with decentralized shields perform comparably to agents with centralized shields in several tasks, allowing shielding to be used in environments with decentralized training and execution for the first time.

Synchronizing decisions across multiple agents in realistic settings is problematic since it requires agents to wait for other agents to terminate and communicate about termination reliably. Ideally, agents should learn and execute asynchronously instead. Such asynchronous methods also allow temporally extended actions that can take different amounts of time based on the situation and action executed. Unfortunately, current policy gradient methods are not applicable in asynchronous settings, as they assume that agents synchronously reason about action selection at every time step. To allow asynchronous learning and decision-making, we formulate a set of asynchronous multi-agent actor-critic methods that allow agents to directly optimize asynchronous policies in three standard training paradigms: decentralized learning, centralized learning, and centralized training for decentralized execution. Empirical results (in simulation and hardware) in a variety of realistic domains demonstrate the superiority of our approaches in large multi-agent problems and validate the effectiveness of our algorithms for learning high-quality and asynchronous solutions.

Offline training in simulated partially observable environments allows reinforcement learning methods to exploit privileged state information through a mechanism known as asymmetry. Such privileged information has the potential to greatly improve the optimal convergence properties, if used appropriately. However, current research in asymmetric reinforcement learning is often heuristic in nature, with few connections to underlying theory or theoretical guarantees, and is primarily tested through empirical evaluation. In this work, we develop the theory of Asymmetric Policy Iteration, an exact model-based dynamic programming solution method, and then apply relaxations which eventually result in Asymmetric DQN, a model-free deep reinforcement learning algorithm. Our theoretical findings are complemented and validated by empirical experimentation performed in environments which exhibit significant amounts of partial observability, and require both information gathering strategies and memorization.

State-of-the-art human-in-the-loop robot grasping is hugely suffered by Electromyography (EMG) inference robustness issues. As a workaround, researchers have been looking into integrating EMG with other signals, often in an ad hoc manner. In this paper, we are presenting a method for end-to-end training of a policy for human-in-the-loop robot grasping on real reaching trajectories. For this purpose we use Reinforcement Learning (RL) and Imitation Learning (IL) in DEXTRON (DEXTerity enviRONment), a stochastic simulation environment with real human trajectories that are augmented and selected using a Monte Carlo (MC) simulation method. We also offer a success model which once trained on the expert policy data and the RL policy roll-out transitions, can provide transparency to how the deep policy works and when it is probably going to fail.

Predictive state representations (PSRs) are models of controlled non-Markov observation sequences which exhibit the same generative process governing POMDP observations without relying on an underlying latent state. In that respect, a PSR is indistinguishable from the corresponding POMDP. However, PSRs notoriously ignore the notion of rewards, which undermines the general utility of PSR models for control, planning, or reinforcement learning. Therefore, we describe a sufficient and necessary accuracy condition which determines whether a PSR is able to accurately model POMDP rewards, we show that rewards can be approximated even when the accuracy condition is not satisfied, and we find that a non-trivial number of POMDPs taken from a well-known third-party repository do not satisfy the accuracy condition. We propose reward-predictive state representations (R-PSRs), a generalization of PSRs which accurately models both observations and rewards, and develop value iteration for R-PSRs. We show that there is a mismatch between optimal POMDP policies and the optimal PSR policies derived from approximate rewards. On the other hand, optimal R-PSR policies perfectly match optimal POMDP policies, reconfirming R-PSRs as accurate state-less generative models of observations and rewards.

We consider the challenges of learning multi-agent/robot macro-action-based deep Q-nets including how to properly update each macro-action value and accurately maintain macro-action-observation trajectories. We address these challenges by first proposing two fundamental frameworks for learning macro-action-value function and joint macro-action-value function. Furthermore, we present two new approaches of learning decentralized macro-action-based policies, which involve a new double Q-update rule that facilitates the learning of decentralized Q-nets by using a centralized Q-net for action selection. Our approaches are evaluated both in simulation and on real robots.

In many real-world multi-robot tasks, high-quality solutions often require a team of robots to perform asynchronous actions under decentralized control. Decentralized multi-agent reinforcement learning methods have difficulty learning decentralized policies because of the environment appearing to be non-stationary due to other agents also learning at the same time. In this paper, we address this challenge by proposing a macro-action-based decentralized multi-agent double deep recurrent Q-net (MacDec-MADDRQN) which trains each decentralized Q-net using a centralized Q-net for action selection. A generalized version of MacDec-MADDRQN with two separate training environments, called Parallel-MacDec-MADDRQN, is also presented to leverage either centralized or decentralized exploration. The advantages and the practical nature of our methods are demonstrated by achieving near-centralized results in simulation and having real robots accomplish a warehouse tool delivery task in an efficient way.

In this paper we investigate how social robots can efficiently gather user preferences without exceeding the allowed user annoyance threshold. To do so, we use a Gazebo based simulated office environment with a TIAGo Steel robot. We then formulate the user annoyance aware preference elicitation problem as a combination of tensor completion and knapsack problems. We then test our approach on the aforementioned simulated environment and demonstrate that it can accurately estimate user preferences.

Collective human knowledge has clearly benefited from the fact that innovations by individuals are taught to others through communication. Similar to human social groups, agents in distributed learning systems would likely benefit from communication to share knowledge and teach skills. The problem of teaching to improve agent learning has been investigated by prior works, but these approaches make assumptions that prevent application of teaching to general multiagent problems, or require domain expertise for problems they can apply to. This learning to teach problem has inherent complexities related to measuring long-term impacts of teaching that compound the standard multiagent coordination challenges. In contrast to existing works, this paper presents the first general framework and algorithm for intelligent agents to learn to teach in a multiagent environment. Our algorithm, Learning to Coordinate and Teach Reinforcement (LeCTR), addresses peer-to-peer teaching in cooperative multiagent reinforcement learning. Each agent in our approach learns both when and what to advise, then uses the received advice to improve local learning. Importantly, these roles are not fixed; these agents learn to assume the role of student and/or teacher at the appropriate moments, requesting and providing advice in order to improve teamwide performance and learning. Empirical comparisons against state-of-the-art teaching methods show that our teaching agents not only learn significantly faster, but also learn to coordinate in tasks where existing methods fail.

The problem of finding and grasping a target object in a cluttered, uncertain environment, target object search, is a common and important problem in robotics. One key challenge is the uncertainty of locating and recognizing each object in a cluttered environment due to noisy perception and occlusions. Furthermore, the uncertainty in localization makes manipulation difficult and uncertain. To cope with these challenges, we formulate the target object search task as a partially observable Markov decision process (POMDP), enabling the robot to reason about perceptual and manipulation uncertainty while searching. To further address the manipulation difficulty, we propose Parameterized Action Partially Observable Monte-Carlo Planning (PA-POMCP), an algorithm that evaluates manipulation actions by taking into account the effect of the robot's current belief on the success of the action execution. In addition, a novel run-time initial belief generator and a state value estimator are introduced in this paper to facilitate the PA-POMCP algorithm. Our experiments show that our methods solve the target object search task in settings where simpler methods either take more object movements or fail.

Modern deep reinforcement learning methods have departed from the incremental learning required for eligibility traces, rendering the implementation of the λ-return difficult in this context. In particular, off-policy methods that utilize experience replay remain problematic because their random sampling of minibatches is not conducive to the efficient calculation of λ-returns. Yet replay-based methods are often the most sample efficient, and incorporating λ-returns into them is a viable way to achieve new state-of-the-art performance. Towards this, we propose the first method to enable practical use of λ-returns in arbitrary replay-based methods without relying on other forms of decorrelation such as asynchronous gradient updates. By promoting short sequences of past transitions into a small cache within the replay memory, adjacent λ-returns can be efficiently precomputed by sharing Q-values. Computation is not wasted on experiences that are never sampled, and stored λ-returns behave as stable temporal-difference (TD) targets that replace the target network. Additionally, our method grants the unique ability to observe TD errors prior to sampling; for the first time, transitions can be prioritized by their true significance rather than by a proxy to it. Furthermore, we propose the novel use of the TD error to dynamically select λ-values that facilitate faster learning. We show that these innovations can enhance the performance of DQN when playing Atari 2600 games, even under partial observability. While our work specifically focuses on λ-returns, these ideas are applicable to any multi-step return estimator.

Decentralized partially observable Markov decision processes (Dec-POMDPs) are general models for decentralized multi-agent decision making under uncertainty. However, they typically model a problem at a low level of granularity, where each agent's actions are primitive operations lasting exactly one time step. We address the case where each agent has macro-actions: temporally extended actions that may require different amounts of time to execute. We model macro-actions as options in a Dec-POMDP, focusing on actions that depend only on information directly available to the agent during execution. Therefore, we model systems where coordination decisions only occur at the level of deciding which macro-actions to execute. The core technical difficulty in this setting is that the options chosen by each agent no longer terminate at the same time. We extend three leading Dec-POMDP algorithms for policy generation to the macro-action case, and demonstrate their effectiveness in both standard benchmarks and a multi-robot coordination problem. The results show that our new algorithms retain agent coordination while allowing high-quality solutions to be generated for significantly longer horizons and larger state-spaces than previous Dec-POMDP methods. Furthermore, in the multi-robot domain, we show that, in contrast to most existing methods that are specialized to a particular problem class, our approach can synthesize control policies that exploit opportunities for coordination while balancing uncertainty, sensor information, and information about other agents.

A key challenge in multi-robot and multi-agent systems is generating solutions that are robust to other self-interested or even adversarial parties who actively try to prevent the agents from achieving their goals. The practicality of existing works addressing this challenge is limited to only small-scale synchronous decision-making scenarios or a single agent planning its best response against a single adversary with fixed, procedurally characterized strategies. In contrast this paper considers a more realistic class of problems where a team of asynchronous agents with limited observation and communication capabilities need to compete against multiple strategic adversaries with changing strategies. This problem necessitates agents that can coordinate to detect changes in adversary strategies and plan the best response accordingly. Our approach first optimizes a set of stratagems that represent these best responses. These optimized stratagems are then integrated into a unified policy that can detect and respond when the adversaries change their strategies. The near-optimality of the proposed framework is established theoretically as well as demonstrated empirically in simulation and hardware.

Multi-agent planning and learning methods are becoming increasingly important in today's interconnected world. Methods for real-world domains, such as robotics, must consider uncertainty and limited communication in order to generate high-quality, robust solutions. This paper discusses our work on developing principled models to represent these problems and planning and learning methods that can scale to realistic multi-agent and multi-robot tasks.

Multiplayer Online Battle Arena (MOBA) games have received increasing popularity recently. In a match of such games, players compete in two teams of five, each controlling an in-game avatar, known as heroes, selected from a roster of more than 100. The selection of heroes, also known as pick or draft, takes place before the match starts and alternates between the two teams until each player has selected one hero. Heroes are designed with different strengths and weaknesses to promote team cooperation in a game. Intuitively, heroes in a strong team should complement each other's strengths and suppress those of opponents. Hero drafting is therefore a challenging problem due to the complex hero-to-hero relationships to consider. In this paper, we propose a novel hero recommendation system that suggests heroes to add to an existing team while maximizing the team's prospect for victory. To that end, we model the drafting between two teams as a combinatorial game and use Monte Carlo Tree Search (MCTS) for estimating the values of hero combinations. Our empirical evaluation shows that hero teams drafted by our recommendation algorithm have a significantly higher win rate against teams constructed by other baseline and state-of-the-art strategies.

The POMDP is a powerful framework for reasoning under outcome and information uncertainty, but constructing an accurate POMDP model is difficult. Bayes-Adaptive Partially Observable Markov Decision Processes (BA-POMDPs) extend POMDPs to allow the model to be learned during execution. BA-POMDPs are a Bayesian RL approach that, in principle, allows for an optimal trade-off between exploitation and exploration. Unfortunately, BA-POMDPs are currently impractical to solve for any non-trivial domain. In this paper, we extend the Monte-Carlo Tree Search method POMCP to BA-POMDPs and show that the resulting method, which we call BA-POMCP, is able to tackle problems that previous solution methods have been unable to solve. Additionally, we introduce several techniques that exploit the BA-POMDP structure to improve the efficiency of BA-POMCP along with proof of their convergence.

Many real-world tasks involve multiple agents with partial observability and limited communication. Learning is challenging in these settings due to local viewpoints of agents, which perceive the world as non-stationary due to concurrently-exploring teammates. Approaches that learn specialized policies for individual tasks face problems when applied to the real world: not only do agents have to learn and store distinct policies for each task, but in practice identities of tasks are often non-observable, making these approaches inapplicable. This paper formalizes and addresses the problem of multi-task multi-agent reinforcement learning under partial observability. We introduce a decentralized single-task learning approach that is robust to concurrent interactions of teammates, and present an approach for distilling single-task policies into a unified policy that performs well across multiple related tasks, without explicit provision of task identity.

This paper presents the first ever approach for solving continuous-observation Decentralized Partially Observable Markov Decision Processes (Dec-POMDPs) and their semi-Markovian counterparts, Dec-POSMDPs. This contribution is especially important in robotics, where a vast number of sensors provide continuous observation data. A continuous-observation policy representation is introduced using Stochastic Kernel-based Finite State Automata (SK-FSAs). An SK-FSA search algorithm titled Entropy-based Policy Search using Continuous Kernel Observations (EPSCKO) is introduced and applied to the first ever continuous-observation Dec-POMDP/Dec-POSMDP domain, where it significantly outperforms state-of-the-art discrete approaches. This methodology is equally applicable to Dec-POMDPs and Dec-POSMDPs, though the empirical analysis presented focuses on Dec-POSMDPs due to their higher scalability. To improve convergence, an entropy injection policy search acceleration approach for both continuous and discrete observation cases is also developed and shown to improve convergence rates without degrading policy quality.

Robust environment perception is essential for decision-making on robots operating in complex domains. Intelligent task execution requires principled treatment of uncertainty sources in a robot's observation model. This is important not only for low-level observations (e.g., accelerom-eter data), but also for high-level observations such as semantic object labels. This paper formalizes the concept of macro-observations in Decentralized Partially Observable Semi-Markov Decision Processes (Dec-POSMDPs), allowing scalable semantic-level multi-robot decision making. A hierarchical Bayesian approach is used to model noise statistics of low-level classifier outputs, while simultaneously allowing sharing of domain noise characteristics between classes. Classification accuracy of the proposed macro-observation scheme, called Hierarchical Bayesian Noise Inference (HBNI), is shown to exceed existing methods. The macro-observation scheme is then integrated into a Dec-POSMDP planner, with hardware experiments running onboard a team of dynamic quadrotors in a challenging domain where noise-agnostic filtering fails. To the best of our knowledge, this is the first demonstration of a real-time, convolutional neural net-based classification framework running fully onboard a team of quadrotors in a multi-robot decision-making domain.

The decentralized partially observable Markov decision process (Dec-POMDP) is a powerful model for representing multi-agent problems with decentralized behavior. Unfortunately, current Dec-POMDP solution methods cannot solve problems with continuous observations, which are common in many real-world domains. To that end, we present a framework for representing and generating Dec-POMDP policies that explicitly include continuous observations. We apply our algorithm to a novel tagging problem and an extended version of a common benchmark, where it generates policies that meet or exceed the values of equivalent discretized domains without the need for finding an adequate discretization.

This paper presents a data-driven approach for multi-robot coordination in partially-observable domains based on Decentralized Partially Observable Markov Decision Processes (Dec-POMDPs) and macro-actions (MAs). Dec-POMDPs provide a general framework for cooperative sequential decision making under uncertainty and MAs allow temporally extended and asynchronous action execution. To date, most methods assume the underlying Dec-POMDP model is known a priori or a full simulator is available during planning time. Previous methods which aim to address these issues suffer from local optimality and sensitivity to initial conditions. Additionally, few hardware demonstrations involving a large team of heterogeneous robots and with long planning horizons exist. This work addresses these gaps by proposing an iterative sampling based Expectation-Maximization algorithm (iSEM) to learn polices using only trajectory data containing observations, MAs, and rewards. Our experiments show the algorithm is able to achieve better solution quality than the state-of-the-art learning-based methods. We implement two variants of multi-robot Search and Rescue (SAR) domains (with and without obstacles) on hardware to demonstrate the learned policies can effectively control a team of distributed robots to cooperate in a partially observable stochastic environment.

Decentralized partially observable Markov decision processes (Dec-POMDPs) provide a general framework for multiagent sequential decision-making under uncertainty. Although Dec-POMDPs are typically intractable to solve for real-world problems, recent research on macro-actions (i.e., temporally-extended actions) has significantly increased the size of problems that can be solved. However, current methods assume the underlying Dec-POMDP model is known a priori or a full simulator is available during planning time. To accommodate more realistic scenarios, when such information is not available, this paper presents a policy-based reinforcement learning approach, which learns the agent policies based solely on trajectories generated by previous interaction with the environment (e.g., demonstrations). We show that our approach is able to generate valid macro-action controllers and develop an expectationmaximization (EM) algorithm (called Policy-based EM or PoEM), which has convergence guarantees for batch learning. Our experiments show PoEM is a scalable learning method that can learn optimal policies and improve upon hand-coded “expert” solutions.

This paper introduces a probabilistic algorithm for multi-robot decision-making under uncertainty, which can be posed as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP). Dec-POMDPs are inherently synchronous decision-making frameworks which require significant computational resources to be solved, making them infeasible for many real-world robotics applications. The Decentralized Partially Observable Semi-Markov Decision Process (Dec-POSMDP) was recently introduced as an extension of the Dec-POMDP that uses high-level macro-actions to allow large-scale, asynchronous decision-making. However, existing Dec-POSMDP solution methods have limited scalability or perform poorly as the problem size grows. This paper proposes a cross-entropy based Dec-POSMDP algorithm motivated by the combinatorial optimization literature. The algorithm is applied to a constrained package delivery domain, where it significantly outperforms existing Dec-POSMDP solution methods.

Decentralized partially observable Markov decision processes (Dec-POMDPs) provide a general model for decision-making under uncertainty in decentralized settings, but are difficult to solve optimally (NEXP-Complete). As a new way of solving these problems, we introduce the idea of transforming a Dec-POMDP into a continuous-state deterministic MDP with a piecewise-linear and convex value function. This approach makes use of the fact that planning can be accomplished in a centralized offline manner, while execution can still be decentralized. This new Dec-POMDP formulation, which we call an occupancy MDP, allows powerful POMDP and continuous-state MDP methods to be used for the first time. To provide scalability, we refine this approach by combining heuristic search and compact representations that exploit the structure present in multi-agent domains, without losing the ability to converge to an optimal solution. In particular, we introduce a feature-based heuristic search value iteration (FB-HSVI) algorithm that relies on feature-based compact representations, point-based updates and efficient action selection. A theoretical analysis demonstrates that FB-HSVI terminates in finite time with an optimal solution. We include an extensive empirical analysis using well-known benchmarks, thereby demonstrating that our approach provides significant scalability improvements compared to the state of the art.

Online, sample-based planning algorithms for POMDPs have shown great promise in scaling to problems with large state spaces, but they become intractable for large action and observation spaces. This is particularly problematic in multiagent POMDPs where the action and observation space grows exponentially with the number of agents. To combat this intractability, we propose a novel scalable approach based on sample-based planning and factored value functions that exploits structure present in many multiagent settings. This approach applies not only in the planning case, but also in the Bayesian reinforcement learning setting. Experimental results show that we are able to provide high quality solutions to large multiagent planning and learning problems.

This paper presents a probabilistic framework for synthesizing control policies for general multi-robot systems that is based on decentralized partially observable Markov decision processes (Dec-POMDPs). Dec-POMDPs are a general model of decision-making where a team of agents must cooperate to optimize a shared objective in the presence of uncertainty. Dec-POMDPs also consider communication limitations, so execution is decentralized. While Dec-POMDPs are typically intractable to solve for real-world problems, recent research on the use of macro-actions in Dec-POMDPs has significantly increased the size of problem that can be practically solved. We show that, in contrast to most existing methods that are specialized to a particular problem class, our approach can synthesize control policies that exploit any opportunities for coordination that are present in the problem, while balancing uncertainty, sensor information, and information about other agents. We use three variants of a warehouse task to show that a single planner of this type can generate cooperative behavior using task allocation, direct communication, and signaling, as appropriate. This demonstrates that our algorithmic framework can automatically optimize control and communication policies for complex multi-robot systems.

The focus of this paper is on solving multi-robot planning problems in continuous spaces with partial observability. Decentralized Partially Observable Markov Decision Processes (Dec-POMDPs) are general models for multi-robot coordination problems, but representing and solving Dec-POMDPs is often intractable for large problems. To allow for a high-level representation that is natural for multi-robot problems and scalable to large discrete and continuous problems, this paper extends the Dec-POMDP model to the Decentralized Partially Observable Semi-Markov Decision Process (Dec-POSMDP). The Dec-POSMDP formulation allows asynchronous decision-making by the robots, which is crucial in multi-robot domains. We also present an algorithm for solving this Dec-POSMDP which is much more scalable than previous methods since it can incorporate closed-loop belief space macro-actions in planning. These macro-actions are automatically constructed to produce robust solutions. The proposed method's performance is evaluated on a complex multi-robot package delivery problem under uncertainty, showing that our approach can naturally represent multi-robot problems and provide high-quality solutions for large-scale problems.

This article presents the state-of-the-art in optimal solution methods for decentralized partially observable Markov decision processes (Dec-POMDPs), which are general models for collaborative multiagent planning under uncertainty. Building off the generalized multiagent A* (GMAA*) algorithm, which reduces the problem to a tree of one-shot collaborative Bayesian games (CBGs), we describe several advances that greatly expand the range of Dec-POMDPs that can be solved optimally. First, we introduce lossless incremental clustering of the CBGs solved by GMAA*, which achieves exponential speedups without sacrificing optimality. Second, we introduce incremental expansion of nodes in the GMAA* search tree, which avoids the need to expand all children, the number of which is in the worst case doubly exponential in the node's depth. This is particularly beneficial when little clustering is possible. In addition, we introduce new hybrid heuristic representations that are more compact and thereby enable the solution of larger Dec-POMDPs. We provide theoretical guarantees that, when a suitable heuristic is used, both incremental clustering and incremental expansion yield algorithms that are both complete and search equivalent. Finally, we present extensive empirical results demonstrating that GMAA*-ICE, an algorithm that synthesizes these advances, can optimally solve Dec-POMDPs of unprecedented size.

With rapid profusion of video data, automated surveillance and intrusion detection is becoming closer to reality. In order to provide timely responses while limiting false alarms, an intrusion detection system must balance resources (e.g., time) and accuracy. In this paper, we show how such a system can be modeled with a partially observable Markov decision process (POMDP), representing possible computer vision filters and their costs in a way that is similar to human vision systems. The POMDP representation can be optimized to produce a dynamic sequence of operations and achieve a tradeoff between time and detection quality, taking into account uncertainty in the filter predictions. In a set of experiments on actual video data, we show that our method can both outperform static “expert” models and scale to large dynamic domains. These results suggest that our method could be used in real-world intrusion detection systems.

Planning under uncertainty for multiagent systems can be formalized as a decentralized partially observable Markov decision process. We advance the state of the art for opti-mal solution of this model, building on the Multiagent A* heuristic search method. A key insight is that we can avoid the full expansion of a search node that generates a number of children doubly exponential in the node’s depth. Instead we incrementally expand the children of a node only when a next child might have the highest heuristic value. We tar-get a subsequent bottleneck by introducing a more memory-efficient representation for our heuristic functions. Proof is given that the resulting algorithm is correct and experiments demonstrate a significant speedup over the state of the art, allowing for optimal solutions over longer horizons for many benchmark problems.

Existing controller-based approaches for centralized and decentralized POMDPs are based on automata with output known as Moore machines. In this paper, we show that several advantages can be gained by utilizing another type of automata, the Mealy machine. Mealy machines are more powerful than Moore machines, provide a richer structure that can be exploited by solution methods, and can be easily incorporated into current controller-based approaches. To demonstrate this, we adapted some existing controller-based algorithms to use Mealy machines and obtained results on a set of benchmark domains. The Mealy-based approach always outperformed the Moore-based approach and often outperformed the state-of-the-art algorithms for both centralized and decentralized POMDPs. These findings provide fresh and general insights for the improvement of existing algorithms and the development of new ones.

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.