Decentralized MDPs with sparse interactions

MOO-MDP: An Object-Oriented Representation for Cooperative Multiagent Reinforcement Learning

Reinforcement learning (RL) is a widely known technique to enable autonomous learning. Even though RL methods achieved successes in increasingly large and complex problems, scaling solutions remains a challenge. One way to simplify (and consequently accelerate) learning is to exploit regularities in a domain, which allows generalization and reduction of the learning space. While object-oriented Markov decision processes (OO-MDPs) provide such generalization opportunities, we argue that the learning process may be further simplified by dividing the workload of tasks amongst multiple agents, solving problems as multiagent systems (MAS). In this paper, we propose a novel combination of OO-MDP and MAS, called multiagent OO-MDP (MOO-MDP). Our proposal accrues the benefits of both OO-MDP and MAS, better addressing scalability issues. We formalize the general model MOO-MDP and present an algorithm to solve deterministic cooperative MOO-MDPs. We show that our algorithm learns optimal policies while reducing the learning space by exploiting state abstractions. We experimentally compare our results with earlier approaches in three domains and evaluate the advantages of our approach in sample efficiency and memory requirements.

Multiagent Reinforcement Learning With Sparse Interactions by Negotiation and Knowledge Transfer

Reinforcement learning has significant applications for multiagent systems, especially in unknown dynamic environments. However, most multiagent reinforcement learning (MARL) algorithms suffer from such problems as exponential computation complexity in the joint state-action space, which makes it difficult to scale up to realistic multiagent problems. In this paper, a novel algorithm named negotiation-based MARL with sparse interactions (NegoSIs) is presented. In contrast to traditional sparse-interaction-based MARL algorithms, NegoSI adopts the equilibrium concept and makes it possible for agents to select the nonstrict equilibrium-dominating strategy profile (nonstrict EDSP) or meta equilibrium for their joint actions. The presented NegoSI algorithm consists of four parts: 1) the equilibrium-based framework for sparse interactions; 2) the negotiation for the equilibrium set; 3) the minimum variance method for selecting one joint action; and 4) the knowledge transfer of local Q -values. In this integrated algorithm, three techniques, i.e., unshared value functions, equilibrium solutions, and sparse interactions are adopted to achieve privacy protection, better coordination and lower computational complexity, respectively. To evaluate the performance of the presented NegoSI algorithm, two groups of experiments are carried out regarding three criteria: 1) steps of each episode; 2) rewards of each episode; and 3) average runtime. The first group of experiments is conducted using six grid world games and shows fast convergence and high scalability of the presented algorithm. Then in the second group of experiments NegoSI is applied to an intelligent warehouse problem and simulated results demonstrate the effectiveness of the presented NegoSI algorithm compared with other state-of-the-art MARL algorithms.

Multiagent Learning of Coordination in Loosely Coupled Multiagent Systems

Multiagent learning (MAL) is a promising technique for agents to learn efficient coordinated behaviors in multiagent systems (MASs). In MAL, concurrent multiple distributed learning processes can make the learning environment nonstationary for each individual learner. Developing an efficient learning approach to coordinate agents' behaviors in this dynamic environment is a difficult problem, especially when agents do not know the domain structure and have only local observability of the environment. In this paper, a coordinated MAL approach is proposed to enable agents to learn efficient coordinated behaviors by exploiting agent independence in loosely coupled MASs. The main feature of the proposed approach is to explicitly quantify and dynamically adapt agent independence during learning so that agents can make a trade-off between a single-agent learning process and a coordinated learning process for an efficient decision making. The proposed approach is employed to solve two-robot navigation problems in different scales of domains. Experimental results show that agents using the proposed approach can learn to act in concert or independently in different areas of the environment, which results in great computational savings and near optimal performance.