Decentralized reinforcement learning of robot behaviors

Virtual soccer simulation is a platform for development of algorithms of multiagent systems. The article describes an approach for development of architecture of agents based on fuzzy states machines, also offers an original mathematical model. The model includes elements of training to increase agility of behavior of the agent. Authors developed general algorithm of fuzzy state machines used in agent to control multiagent systems with it. In addition, they proposed software structure for agile interaction between state machines and soccer simulation server and for storing results of training assets.

Application of Fuzzy State Machines to Control Players in Virtual Soccer Simulation

Reinforcement learning is a subfield of machine learning which is similar to human learning. In recent years, it has drawn a considerable portion of researchers' attention and it has been revolutionized; such as its integration with deep learning. This integration has created a better understanding of the visual environments and end-to-end direct learning from pixels to solve problems that have previously been intractable. This improvement has led to the creation of various deep reinforcement learning algorithms with different goals. In this paper, deep reinforcement learning algorithms and their applications are reviewed and categorized. This work also addresses the advantages and disadvantages of algorithms and the challenges that are solved with appearance of deep reinforcement learning. In order to use these algorithms, there are important considerations that need to be addressed in each problem. These considerations are about the most important components of reinforcement learning, which has been analyzed and categorized as the important achievement of this paper.

Deep Reinforcement Learning Issues and Approaches for The Multi-Agent Centric Problems

/pdf/a-survey-on-privacy-for-b5g-6g-new-privacy-challenges-and-1l7r15tr.pdf

A survey on privacy for B5G/6G: New privacy challenges, and research directions

Accelerating decentralized reinforcement learning of complex individual behaviors

In this paper, decentralized reinforcement learning is applied to a control problem with a multidimensional action space. We propose a decentralized reinforcement learning architecture for a mobile robot, where the individual components of the commanded velocity vector are learned in parallel by separate agents. We empirically demonstrate that the decentralized architecture outperforms its centralized counterpart in terms of the learning time, while using less computational resources. The method is validated on two problems: an extended version of the 3-dimensional mountain car, and a ball-pushing behavior performed with a differential-drive robot, which is also tested on a physical setup.

Aashish Vatsyayan

Papers

Decentralized Reinforcement Learning Applied to Mobile Robots