Design of semi-decentralized control laws for distributed-air-jet micromanipulators by reinforcement learning

Abstract

Design of semi-decentralized control laws for distributed-air-jet micromanipulators by reinforcement learning Laëtitia Matignon, Guillaume J. Laurent and Nadine Le Fort-PiatAbstract-Recently, a great deal of interest has been developed in learning in multi-agent systems to achieve decentralized control. Machine learning is a popular approach to find controllers that are tailored exactly to the system without any prior model. In this paper, we propose a semi-decentralized reinforcement learning control approach in order to position and convey an object on a contact-free MEMS-based distributedmanipulation system. The experimental results validate the semi-decentralized reinforcement learning method as a way to design control laws for such distributed systems.

Decentralized Reinforcement Learning of Robot Behaviors

“…Kimura [20] presents a coarse coding technique and an action selection scheme for RL in multi-dimensional and continuous state-action spaces following conventional and sound RL manners; and Pazis and Lagoudakis [38] present an approach for efficiently learning and acting in domains with continuous and/or multidimensional control variables, in which the problem of generalizing among actions is transformed to a problem of generalizing among states in an equivalent MDP, where action selection is trivial. A different application is reported by Matignon, Laurent, and Fort-Piat [32], where a semi-decentralized RL control approach for controlling a distributed-air-jet micro-manipulator is proposed. This showed a successful application of decentralized Q-learning variant algorithms for independent agents.…”

Distributed part differentiation in a smart surface

“…The goal of the Smart Surface project (see [13,14]) is to design a distributed and integrated micromanipulator based on an array of micromodules or cells and to develop an automated positioning and conveying surface. Each micromodule will be composed of a microactuator, a microsensor and a processing unit.…”

Distributed Discrete State Acquisition and Concurrent Pattern Recognition in a MEMS-Based Smart Surface#

“…The objective of the Smart Surface proje [18]) is to design a distributed and in manipulator based on an array of micro mod realize an automated positioning and conveyi micro module will be composed of a micro a sensor and a processing unit. The coope modules thanks to an integrated network will to recognize the parts and to control micro actu move and position accurately the parts on th We consider small parts that cover a small n modules and that are moved via air nozzle rectangular holes on the front-side of Fig.…”