Training Unity Machine Learning Agents using reinforcement learning method

Reinforcement learning (RL) is a kind of machine learning. It aims to optimize agents’ policies by adapting the agents to an environment according to rewards. In this paper, we propose a method for improving policies by using stochastic knowledge, in which reinforcement learning agents obtain. We use a Bayesian Network (BN), which is a stochastic model, as knowledge of an agent. Its structure is decided by minimum description length criterion using series of an agent’s input-output and rewards as sample data. A BN constructed in our study represents stochastic dependences between input-output and rewards. In our proposed method, policies are improved by supervised learning using the structure of BN (i.e. stochastic knowledge). The proposed improvement mechanism makes RL agents acquire more effective policies. We carry out simulations in the pursuit problem in order to show the effectiveness of our proposed method.

Download Full-text

THE ASSIGNMENT OF TASKS TO THE NODES OF THE IOT DISTRIBUTED SYSTEM BASED ON REINFORCEMENT MACHINE LEARNING

Автоматизация процессов управления ◽

10.35752/1991-2927-2021-1-63-27-33 ◽

2021 ◽

Vol 63 (1) ◽

pp. 27-33

Author(s):

Maria V. Stepanova ◽

◽

Oleg I. Eremin ◽

Keyword(s):

Machine Learning ◽

Reinforcement Learning ◽

Internet Of Things ◽

Distributed System ◽

Task Assignment ◽

Parallel Systems ◽

Learning Method ◽

Adaptive Approach ◽

Iot Platform

The article describes issues of applying an adaptive approach based on reinforcement learning for assignment of the computing tasks to nodes of distributed Internet of Things (IoT) platform. The IoT platform consists of heterogeneous elements that are computing nodes. Classical approaches, methods, and algorithms for distributed and parallel systems are not suitable for task assignment in IoT systems due to its characteristics. The reinforcement learning method allows you to solve the problem of building a distributed system due to the adaptive formation of a sequence of computational nodes and the corresponding computational tasks. Thus, the article represents a method that makes IoT nodes capable of execution computing tasks, especially, which were previously designed for classical distributed and parallel systems.

Download Full-text

Learning and generalising object extraction skill for contact-rich disassembly tasks: an introductory study

The International Journal of Advanced Manufacturing Technology ◽

10.1007/s00170-021-08086-z ◽

2021 ◽

Author(s):

Antonio Serrano-Muñoz ◽

Nestor Arana-Arexolaleiba ◽

Dimitrios Chrysostomou ◽

Simon Bøgh

Keyword(s):

Machine Learning ◽

Reinforcement Learning ◽

State Of The Art ◽

Learning Algorithms ◽

Robotic Manipulation ◽

Object Extraction ◽

Learning Agents ◽

Machine Learning Methods ◽

Key Concepts ◽

Planning And Operation

AbstractRemanufacturing automation must be designed to be flexible and robust enough to overcome the uncertainties, conditions of the products, and complexities in the planning and operation of the processes. Machine learning methods, in particular reinforcement learning, are presented as techniques to learn, improve, and generalise the automation of many robotic manipulation tasks (most of them related to grasping, picking, or assembly). However, not much has been exploited in remanufacturing, in particular in disassembly tasks. This work presents the state of the art of contact-rich disassembly using reinforcement learning algorithms and a study about the generalisation of object extraction skills when applied to contact-rich disassembly tasks. The generalisation capabilities of two state-of-the-art reinforcement learning agents (trained in simulation) are tested and evaluated in simulation, and real world while perform a disassembly task. Results show that at least one of the agents can generalise the contact-rich extraction skill. Besides, this work identifies key concepts and gaps for the reinforcement learning algorithms’ research and application on disassembly tasks.

Download Full-text

Learning and Generalising Object Extraction Skill for Contact-rich Disassembly Tasks: An Introductory Study

10.21203/rs.3.rs-331448/v1 ◽

2021 ◽

Author(s):

Antonio Serrano Muñoz ◽

Nestor Arana-Arexolaleiba ◽

Dimitrios Chrysostomou ◽

Simon Bøgh

Keyword(s):

Machine Learning ◽

Reinforcement Learning ◽

Real World ◽

State Of The Art ◽

Learning Algorithms ◽

Robotic Manipulation ◽

Object Extraction ◽

Learning Agents ◽

Key Concepts ◽

Planning And Operation

Abstract Remanufacturing automation must be designed to be flexible and robust enough to overcome the uncertainties, conditions of the products, and complexities in the process's planning and operation. Machine learning, particularly reinforcement learning, methods are presented as techniques to learn, improve, and generalise the automation of many robotic manipulation tasks (most of them related to grasping, picking, or assembly). However, not much has been exploited in remanufacturing, in particular in disassembly tasks. This work presents the State-of-the-Art of contact-rich disassembly using reinforcement learning algorithms and a study about the object extraction skill's generalisation when applied to contact-rich disassembly tasks. The generalisation capabilities of two State-of-the-Art reinforcement learning agents (trained in simulation) are tested and evaluated in simulation and real-world while perform a disassembly task. Results shows that, at least, one of the agents can generalise the contact-rich extraction skill. Also, this work identifies key concepts and gaps for the reinforcement learning algorithms' research and application on disassembly tasks.

Download Full-text

Report on the 2008 Reinforcement Learning Competition

AI Magazine ◽

10.1609/aimag.v31i2.2227 ◽

2010 ◽

Vol 31 (2) ◽

pp. 81 ◽

Cited By ~ 19

Author(s):

Shimon Whiteson ◽

Brian Tanner ◽

Adam White

Keyword(s):

Machine Learning ◽

Reinforcement Learning ◽

Empirical Evaluation ◽

The Novel ◽

Software Infrastructure ◽

Robust Learning ◽

Learning Agents ◽

International Conference ◽

The World ◽

History Of

This article reports on the 2008 Reinforcement Learning Competition, which began in November 2007 and ended with a workshop at the International Conference on Machine Learning (ICML) in July 2008 in Helsinki, Finland. Researchers from around the world developed reinforcement learning agents to compete in six problems of various complexity and difficulty. The competition employed fundamentally redesigned evaluation frameworks that, unlike those in previous competitions, aimed to systematically encourage the submission of robust learning methods. We describe the unique challenges of empirical evaluation in reinforcement learning and briefly review the history of the previous competitions and the evaluation frameworks they employed. We also describe the novel frameworks developed for the 2008 competition as well as the software infrastructure on which they rely. Furthermore, we describe the six competition domains and present a summary of selected competition results. Finally, we discuss the implications of these results and outline ideas for the future of the competition.

Download Full-text