Some researchers have speculated that capable reinforcement learning (RL) agents pursuing misspecified objectives are often incentivized to seek resources and power in pursuit of those objectives. An agent seeking power is incentivized to behave in undesirable ways, including rationally preventing deactivation and correction. Others have voiced skepticism: humans seem idiosyncratic in their urges to power, which need not be present in the agents we design. We formalize a notion of power within the context of finite deterministic Markov decision processes (MDPs). We prove that, with respect to a wide class of reward function distributions, optimal policies tend to seek power over the environment.

To achieve these common goals, agents in a MAS should be capable of interacting with other agents, not simply by exchanging data, but by engaging as in social activities, such as those people participate in their daily lives: cooperation, coordination, negotiation, and the like. In MASs, agents are assumed to be autonomous - capable of making independent decisions about to do in order to satisfy their design objectives, and thus they need mechanisms that allow them to synchronize and to coordinate their activities at run time [31]. Although one of the main issues in MASs is the agents' coordination structure, this is not hard-wired at design time, as MASs are typically in standard concurrent/distributed systems. One well-known strategy for coordination in MAS is the design of multi-agent coordinated plans [7][35][36][33][14] that include, not only usual agents' actions defined by their effectors, but also communication actions to achieve the necessary synchronization and coordination. To represent communication actions, some specific languages were created, e.g.

This article reviews recent advances in multi-agent reinforcement learning algorithms for large-scale control systems and communication networks, which learn to communicate and cooperate. We provide an overview of this emerging field, with an emphasis on the decentralized setting under different coordination protocols. We highlight the evolution of reinforcement learning algorithms from single-agent to multi-agent systems, from a distributed optimization perspective, and conclude with future directions and challenges, in the hope to catalyze the growing synergy among distributed optimization, signal processing, and reinforcement learning communities.

In this paper, we study the one-shot and lifelong versions of the Target Assignment and Path Finding problem in automated sortation centers, where each agent needs to constantly assign itself a sorting station, move to its assigned station without colliding with obstacles or other agents, wait in the queue of that station to obtain a parcel for delivery, and then deliver the parcel to a sorting bin. The throughput of such centers is largely determined by the total idle time of all stations since their queues can frequently become empty. To address this problem, we first formalize and study the one-shot version that assigns stations to a set of agents and finds collision-free paths for the agents to their assigned stations. We present efficient algorithms for this task based on a novel min-cost max-flow formulation that minimizes the total idle time of all stations in a fixed time window. We then demonstrate how our algorithms for solving the one-shot problem can be applied to solving the lifelong problem as well. Experimentally, we believe to be the first researchers to consider real-world automated sortation centers using an industrial simulator with realistic data and a kinodynamic model of real robots.

In this paper, we investigate learning temporal abstractions in cooperative multi-agent systems using the options framework (Sutton et al, 1999) and provide a model-free algorithm for this problem. First, we address the planning problem for the decentralized POMDP represented by the multi-agent system, by introducing a common information approach. We use common beliefs and broadcasting to solve an equivalent centralized POMDP problem. Then, we propose the Distributed Option Critic (DOC) algorithm, motivated by the work of Bacon et al (2017) in the single-agent setting. Our approach uses centralized option evaluation and decentralized intra-option improvement. We analyze theoretically the asymptotic convergence of DOC and validate its performance in grid-world environments, where we implement DOC using a deep neural network. Our experiments show that DOC performs competitively with state-of-the-art algorithms and that it is scalable when the number of agents increases.

In large-scale multi-agent systems, the large number of agents and complex game relationship cause great difficulty for policy learning. Therefore, simplifying the learning process is an important research issue. In many multi-agent systems, the interactions between agents often happen locally, which means that agents neither need to coordinate with all other agents nor need to coordinate with others all the time. Traditional methods attempt to use pre-defined rules to capture the interaction relationship between agents. However, the methods cannot be directly used in a large-scale environment due to the difficulty of transforming the complex interactions between agents into rules. In this paper, we model the relationship between agents by a complete graph and propose a novel game abstraction mechanism based on two-stage attention network (G2ANet), which can indicate whether there is an interaction between two agents and the importance of the interaction. We integrate this detection mechanism into graph neural network-based multi-agent reinforcement learning for conducting game abstraction and propose two novel learning algorithms GA-Comm and GA-AC. We conduct experiments in Traffic Junction and Predator-Prey. The results indicate that the proposed methods can simplify the learning process and meanwhile get better asymptotic performance compared with state-of-the-art algorithms.

Recent years have witnessed significant advances in reinforcement learning (RL), which has registered great success in solving various sequential decision-making problems in machine learning. Most of the successful RL applications, e.g., the games of Go and Poker, robotics, and autonomous driving, involve the participation of more than one single agent, which naturally fall into the realm of multi-agent RL (MARL), a domain with a relatively long history, and has recently re-emerged due to advances in single-agent RL techniques. Though empirically successful, theoretical foundations for MARL are relatively lacking in the literature. In this chapter, we provide a selective overview of MARL, with focus on algorithms backed by theoretical analysis. More specifically, we review the theoretical results of MARL algorithms mainly within two representative frameworks, Markov/stochastic games and extensive-form games, in accordance with the types of tasks they address, i.e., fully cooperative, fully competitive, and a mix of the two. We also introduce several significant but challenging applications of these algorithms. Orthogonal to the existing reviews on MARL, we highlight several new angles and taxonomies of MARL theory, including learning in extensive-form games, decentralized MARL with networked agents, MARL in the mean-field regime, (non-)convergence of policy-based methods for learning in games, etc. Some of the new angles extrapolate from our own research endeavors and interests. Our overall goal with this chapter is, beyond providing an assessment of the current state of the field on the mark, to identify fruitful future research directions on theoretical studies of MARL. We expect this chapter to serve as continuing stimulus for researchers interested in working on this exciting while challenging topic.

Multi-agent coordination is prevalent in many real-world applications. However, such coordination is challenging due to its combinatorial nature. An important observation in this regard is that agents in the real world often only directly affect a limited set of neighboring agents. Leveraging such loose couplings among agents is key to making coordination in multi-agent systems feasible. In this work, we focus on learning to coordinate. Specifically, we consider the multi-agent multi-armed bandit framework, in which fully cooperative loosely-coupled agents must learn to coordinate their decisions to optimize a common objective. As opposed to in the planning setting, for learning methods it is challenging to establish theoretical guarantees. We propose multi-agent Thompson sampling (MATS), a new Bayesian exploration-exploitation algorithm that leverages loose couplings. We provide a regret bound that is sublinear in time and low-order polynomial in the highest number of actions of a single agent for sparse coordination graphs. Finally, we empirically show that MATS outperforms the state-of-the-art algorithm, MAUCE, on two synthetic benchmarks, a realistic wind farm control task, and a novel benchmark with Poisson distributions.

We consider the facility location problem in the one-dimensional setting where each facility can serve a limited number of agents from the algorithmic and mechanism design perspectives. From the algorithmic perspective, we prove t hat the corresponding optimization problem, where the goal is t o locate facilities to minimize either the total cost to all ag ents or the maximum cost of any agent is NPhard. However, we show that the problem is fixed-parameter tractable, and the optimal solution can be computed in polynomial time whenever the number of facilities is bounded, or when all facilit ies have identical capacities. We then consider the problem fro m a mechanism design perspective where the agents are strategic and need not reveal their true locations. We show that sev - eral natural mechanisms studied in the uncapacitated setti ng either lose strategyproofness or a bound on the solution qua l-ity for the total or maximum cost objective. We then propose new mechanisms that are strategyproof and achieve approximation guarantees that almost match the lower bounds.

Existing value-factorized based Multi-Agent deep Reinforce-ment Learning (MARL) approaches are well-performing invarious multi-agent cooperative environment under thecen-tralized training and decentralized execution(CTDE) scheme,where all agents are trained together by the centralized valuenetwork and each agent execute its policy independently. How-ever, an issue remains open: in the centralized training process,when the environment for the team is partially observable ornon-stationary, i.e., the observation and action informationof all the agents cannot represent the global states, existingmethods perform poorly and sample inefficiently. Regret Min-imization (RM) can be a promising approach as it performswell in partially observable and fully competitive settings.However, it tends to model others as opponents and thus can-not work well under the CTDE scheme. In this work, wepropose a novel team RM based Bayesian MARL with threekey contributions: (a) we design a novel RM method to traincooperative agents as a team and obtain a team regret-basedpolicy for that team; (b) we introduce a novel method to de-compose the team regret to generate the policy for each agentfor decentralized execution; (c) to further improve the perfor-mance, we leverage a differential particle filter (a SequentialMonte Carlo method) network to get an accurate estimation ofthe state for each agent. Experimental results on two-step ma-trix games (cooperative game) and battle games (large-scalemixed cooperative-competitive games) demonstrate that ouralgorithm significantly outperforms state-of-the-art methods.