In cooperative multi-agent reinforcement learning (MARL), combining value decomposition with actor-critic enables agents to learn stochastic policies, which are more suitable for the partially observable environment. Given the goal of learning local policies that enable decentralized execution, agents are commonly assumed to be independent of each other, even in centralized training. However, such an assumption may prohibit agents from learning the optimal joint policy. To address this problem, we explicitly take the dependency among agents into centralized training. Although this leads to the optimal joint policy, it may not be factorized for decentralized execution. Nevertheless, we theoretically show that from such a joint policy, we can always derive another joint policy that achieves the same optimality but can be factorized for decentralized execution. To this end, we propose multi-agent conditional policy factorization (MACPF), which takes more centralized training but still enables decentralized execution. We empirically verify MACPF in various cooperative MARL tasks and demonstrate that MACPF achieves better performance or faster convergence than baselines. Our code is available at https://github.com/PKU-RL/FOP-DMAC-MACPF.

Abstract--Multi-Agent Reinforcement Learning (MARL) is a promising area of research that can model and control multiple, autonomous decision-making agents. During online training, MARL algorithms involve performance-intensive computations such as exploration and exploitation phases originating from large observation-action space belonging to multiple agents. In this article, we seek to characterize the scalability bottlenecks in several popular classes of MARL algorithms during their training phases. Our experimental results reveal new insights into the key modules of MARL algorithms that limit the scalability, and outline potential strategies that may help address these performance issues. Distributed Artificial Intelligence < I.2 Artificial Intelligence < I Computing Methodologies The function that determines the action is known as a policy.

The research community has shown increasing interest in designing intelligent embodied agents that can assist humans in accomplishing tasks. Despite recent progress on related vision-language benchmarks, most prior work has focused on building agents that follow instructions rather than endowing agents the ability to ask questions to actively resolve ambiguities arising naturally in embodied environments. To empower embodied agents with the ability to interact with humans, in this work, we propose an Embodied Learning-By-Asking (ELBA) model that learns when and what questions to ask to dynamically acquire additional information for completing the task. We evaluate our model on the TEACH vision-dialog navigation and task completion dataset. Experimental results show that ELBA achieves improved task performance compared to baseline models without question-answering capabilities.

Training a team to complete a complex task via multi-agent reinforcement learning can be difficult due to challenges such as policy search in a large policy space, and non-stationarity caused by mutually adapting agents. To facilitate efficient learning of complex multi-agent tasks, we propose an approach which uses an expert-provided curriculum of simpler multi-agent sub-tasks. In each sub-task of the curriculum, a subset of the entire team is trained to acquire sub-task-specific policies. The sub-teams are then merged and transferred to the target task, where their policies are collectively fined tuned to solve the more complex target task. We present MEDoE, a flexible method which identifies situations in the target task where each agent can use its sub-task-specific skills, and uses this information to modulate hyperparameters for learning and exploration during the fine-tuning process. We compare MEDoE to multi-agent reinforcement learning baselines that train from scratch in the full task, and with na\"ive applications of standard multi-agent reinforcement learning techniques for fine-tuning. We show that MEDoE outperforms baselines which train from scratch or use na\"ive fine-tuning approaches, requiring significantly fewer total training timesteps to solve a range of complex teamwork tasks.

To enable space mission sets like on-orbit servicing and manufacturing, agents in close proximity maybe operating too close to yield resolved localization solutions to operators from ground sensors. This leads to a requirement on the systems need to maintain a catalog of their local neighborhood, however, this may impose a large burden on each agent by requiring updating and maintenance of this catalog at each node. To alleviate this burden, this paper considers the case of a single satellite agent (a chief) updating a single catalog. More specifically, we consider the case of numerous satellite deputy agents in a local neighborhood of a chief, the goal of the chief satellite is to maintain and update a catalog of all agents within this neighborhood through onboard measurements. We consider the agents having relative translational and attitude motion dynamics between the chief and deputy, with the chief centered at the origin of the frame. We provide an end-to-end solution of the this problem through providing both a supervisory control method coupled with a Bayesian Filter that propagates the belief state and provides the catalog solutions to the supervisor. The goal of the supervisory controller is to determine which agent to look at and at which times while adhering to constraints of the chief satellite. We provide a numerical validation to this problem with three agents.

In iterative approaches to empirical game-theoretic analysis (EGTA), the strategy space is expanded incrementally based on analysis of intermediate game models. A common approach to strategy exploration, represented by the double oracle algorithm, is to add strategies that best-respond to a current equilibrium. This approach may suffer from overfitting and other limitations, leading the developers of the policy-space response oracle (PSRO) framework for iterative EGTA to generalize the target of best response, employing what they term meta-strategy solvers (MSSs). Noting that many MSSs can be viewed as perturbed or approximated versions of Nash equilibrium, we adopt an explicit regularization perspective to the specification and analysis of MSSs. We propose a novel MSS called regularized replicator dynamics (RRD), which simply truncates the process based on a regret criterion. We show that RRD is more adaptive than existing MSSs and outperforms them in various games. We extend our study to three-player games, for which the payoff matrix is cubic in the number of strategies and so exhaustively evaluating profiles may not be feasible. We propose a profile search method that can identify solutions from incomplete models, and combine this with iterative model construction using a regularized MSS. Finally, and most importantly, we reveal that the regret of best response targets has a tremendous influence on the performance of strategy exploration through experiments, which provides an explanation for the effectiveness of regularization in PSRO.

In recent years, reinforcement learning and its multi-agent analogue have achieved great success in solving various complex control problems. However, multi-agent reinforcement learning remains challenging both in its theoretical analysis and empirical design of algorithms, especially for large swarms of embodied robotic agents where a definitive toolchain remains part of active research. We use emerging state-of-the-art mean-field control techniques in order to convert many-agent swarm control into more classical single-agent control of distributions. This allows profiting from advances in single-agent reinforcement learning at the cost of assuming weak interaction between agents. However, the mean-field model is violated by the nature of real systems with embodied, physically colliding agents. Thus, we combine collision avoidance and learning of mean-field control into a unified framework for tractably designing intelligent robotic swarm behavior. On the theoretical side, we provide novel approximation guarantees for general mean-field control both in continuous spaces and with collision avoidance. On the practical side, we show that our approach outperforms multi-agent reinforcement learning and allows for decentralized open-loop application while avoiding collisions, both in simulation and real UAV swarms. Overall, we propose a framework for the design of swarm behavior that is both mathematically well-founded and practically useful, enabling the solution of otherwise intractable swarm problems.

Task allocation is an important problem for robot swarms to solve, allowing agents to reduce task completion time by performing tasks in a distributed fashion. Existing task allocation algorithms often assume prior knowledge of task location and demand or fail to consider the effects of the geometric distribution of tasks on the completion time and communication cost of the algorithms. In this paper, we examine an environment where agents must explore and discover tasks with positive demand and successfully assign themselves to complete all such tasks. We first provide a new discrete general model for modeling swarms. Operating within this theoretical framework, we propose two new task allocation algorithms for initially unknown environments -- one based on N-site selection and the other on virtual pheromones. We analyze each algorithm separately and also evaluate the effectiveness of the two algorithms in dense vs. sparse task distributions. Compared to the Levy walk, which has been theorized to be optimal for foraging, our virtual pheromone inspired algorithm is much faster in sparse to medium task densities but is communication and agent intensive. Our site selection inspired algorithm also outperforms Levy walk in sparse task densities and is a less resource-intensive option than our virtual pheromone algorithm for this case. Because the performance of both algorithms relative to random walk is dependent on task density, our results shed light on how task density is important in choosing a task allocation algorithm in initially unknown environments.

Achieving distributed reinforcement learning (RL) for large-scale cooperative multi-agent systems (MASs) is challenging because: (i) each agent has access to only limited information; (ii) issues on convergence or computational complexity emerge due to the curse of dimensionality. In this paper, we propose a general computationally efficient distributed framework for cooperative multi-agent reinforcement learning (MARL) by utilizing the structures of graphs involved in this problem. We introduce three coupling graphs describing three types of inter-agent couplings in MARL, namely, the state graph, the observation graph and the reward graph. By further considering a communication graph, we propose two distributed RL approaches based on local value-functions derived from the coupling graphs. The first approach is able to reduce sample complexity significantly under specific conditions on the aforementioned four graphs. The second approach provides an approximate solution and can be efficient even for problems with dense coupling graphs. Here there is a trade-off between minimizing the approximation error and reducing the computational complexity. Simulations show that our RL algorithms have a significantly improved scalability to large-scale MASs compared with centralized and consensus-based distributed RL algorithms.

Why should moral philosophers, moral psychologists, and machine ethicists care about computational complexity? Debates on whether artificial intelligence (AI) can or should be used to solve problems in ethical domains have mainly been driven by what AI can or cannot do in terms of human capacities. In this paper, we tackle the problem from the other end by exploring what kind of moral machines are possible based on what computational systems can or cannot do. To do so, we analyze normative ethics through the lens of computational complexity. First, we introduce computational complexity for the uninitiated reader and discuss how the complexity of ethical problems can be framed within Marr's three levels of analysis. We then study a range of ethical problems based on consequentialism, deontology, and virtue ethics, with the aim of elucidating the complexity associated with the problems themselves (e.g., due to combinatorics, uncertainty, strategic dynamics), the computational methods employed (e.g., probability, logic, learning), and the available resources (e.g., time, knowledge, learning). The results indicate that most problems the normative frameworks pose lead to tractability issues in every category analyzed. Our investigation also provides several insights about the computational nature of normative ethics, including the differences between rule- and outcome-based moral strategies, and the implementation-variance with regard to moral resources. We then discuss the consequences complexity results have for the prospect of moral machines in virtue of the trade-off between optimality and efficiency. Finally, we elucidate how computational complexity can be used to inform both philosophical and cognitive-psychological research on human morality by advancing the Moral Tractability Thesis (MTT).