Value function factorization with dynamic weighting for deep multi-agent reinforcement learning

Du, Wei; Ding, Shifei; Guo, Lili; Zhang, Jian; Zhang, Chenglong; Ding, Ling

doi:10.1016/j.ins.2022.10.042

Cited by 8 publications

(4 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…𝑉 𝑖 (𝑠, 𝜋 1 , … , 𝜋 𝑗 , … , 𝜋 𝑁 ) = ∑ 𝑉 𝜋 (0) 𝑁 𝑗=1 (15) In the Markov model, if the expected reward 𝑉 𝑗 (𝜋…”

Section: F Task Planning Model Based On Markov Gamementioning

confidence: 99%

See 1 more Smart Citation

MARL-Based Multi-Satellite Intelligent Task Planning Method

Zhang,

Li,

et al. 2023

IEEE Access

View full text Add to dashboard Cite

In this article, we propose a solution to multi-satellite intelligent task planning using the multiagent reinforcement learning (MARL) method. Fristly, we have developed a multi-satellite task planning model based on the Markov game framework. Furthermore, we have computationally designed a satellite state transition function to address the task planning problem and successfully solved it using the multi-agent proximal policy optimization (MAPPO) algorithm. Our experimental results demonstrate that the MARL method exhibits remarkable convergence speed and performance, delivering significant rewards in multiscale task planning scenarios. Consequently, it proves to be a highly suitable approach for multi-satellite intelligent task planning.INDEX TERMS MARL; multi-satellite intelligent task planning; Markov game; MAPPO

show abstract

“…𝑉 𝑖 (𝑠, 𝜋 1 , … , 𝜋 𝑗 , … , 𝜋 𝑁 ) = ∑ 𝑉 𝜋 (0) 𝑁 𝑗=1 (15) In the Markov model, if the expected reward 𝑉 𝑗 (𝜋…”

Section: F Task Planning Model Based On Markov Gamementioning

confidence: 99%

“…Littman [14] introduced the MARL approach in the 1990s, using Markov Decision Processes (MDP) as the framework for environment modeling. MARL provides a mathematical framework for solving various reinforcement learning problems and has become the foundation for subsequent studies [15] .…”

Section: Introductionmentioning

confidence: 99%

MARL-Based Multi-Satellite Intelligent Task Planning Method

Zhang,

Li,

et al. 2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…MARL can be used to optimize collaborative detection. Compared with RL, one agent's action will also become the state of other agents after it takes actions according to the strategy in MARL, which will lead to complex state space and convergence difficulty [25]. One of the important research contents of MARL is how to optimize the behavior of each agent without explicit communication among agents.…”

Section: Introductionmentioning

confidence: 99%

An Optimization Method for Collaborative Radar Antijamming Based on Multi-Agent Reinforcement Learning

Feng

Wang

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Attacking a naval vessel with multiple missiles is an important way to improve the hit rate of missiles. Missile-borne radars need to complete detection and antijamming tasks to guide missiles, but communication between these radars is often difficult. In this paper, an optimization method based on multi-agent reinforcement learning is proposed for the collaborative detection and antijamming tasks of multiple radars against one naval vessel. We consider the collaborative radars as one player to make their confrontation with the naval vessel a two-person zero-sum game. With temporal constraints of the radar’s and jammer’s recognition and preparation interval, the game focuses on taking a favorable position at the end of the confrontation. It is assumed the total jamming capability of a shipborne jammer is constant and limited, and the shipborne jammer allocates the jamming capability in the radar’s direction according to the radar threat assessment result and its probability of successful detection. The radars work collaboratively through prior centralized training and obtain a good performance by decentralized execution. The proposed method can make radars collaborate to detect the naval vessel, rather than only considering the detection result of each radar itself. Experimental results show that the proposed method in this paper is effective, improving the winning probability to 10% and 25% in the two-radar and four-radar scenarios, respectively.

show abstract

“…However, to create functionalities with the help of machine learning and deep learning for build such complex and hard-coded systems. In this path planner, reinforcement learning algorithms are used to determine the optimal control sequence on the interaction with the environment during learning, a technique within machine learning [2]. This proposed method is based on reinforcement learning algorithms to model the "ships' motion", and "turning rate".…”

Section: Introductionmentioning

confidence: 99%

Identify the Dynamic Position of Moving Vessels using Big Data and Deep Learning Approaches

Mistry¹,

Suaibia²,

Bin³

2023

EJECE

View full text Add to dashboard Cite

Recently, various purposes such as computer vision, self-driving vehicles, self-directed shipping, and so on. Therefore, machine learning and deep learning algorithms have demonstrated strong routine. Using machine learning applications, we can implement sophisticated functionality without relying on complex code constructions. An alternative is to train a learning system on a collected training dataset and ensure that it performs as expected. Two main benefits of a deep-learning-based system over a hard-coded one exist. A Reinforcement Learning, and deep learning-based system does not require such complex hard-coded algorithms, which makes it less prone to error and easier to implement. In this study we proposed a deep learning-driven systems are able to adapt to different situations by retraining their algorithms on collected data. Proposed method influenced varying obtain leaning to another example of changing conditions. Systems can adapt even when input distributions change over time.

show abstract

Value function factorization with dynamic weighting for deep multi-agent reinforcement learning

Cited by 8 publications

References 17 publications

MARL-Based Multi-Satellite Intelligent Task Planning Method

MARL-Based Multi-Satellite Intelligent Task Planning Method

An Optimization Method for Collaborative Radar Antijamming Based on Multi-Agent Reinforcement Learning

Identify the Dynamic Position of Moving Vessels using Big Data and Deep Learning Approaches

Contact Info

Product

Resources

About