The paper presents a universal architectural pattern and an associated specification method that can be applied in the design of robot control systems. The approach a describes the system in terms of embodied agents and proposes a multi-step decomposition enabling precise definition of their inner structure and operation. An embodied agent is decomposed into effectors, receptors, both real and virtual, and a control subsystem. Those entities communicate through communication buffers. The activities of those entities are governed by FSMs that invoke behaviours formulated in terms of transition functions taking as arguments the contents of input buffers and producing the values inserted into output buffers. The method is exemplified by applying it to the design of a control system of a robot executing one of the most important tasks for a service robot, i.e. picking up, by a position-force controlled robot, an object located using an RGB-D image acquired from a Kinect. Moreover in order to substantiate the universality of the presented approach we present how classical, known from the literature, robotic architectures can be expressed as systems composed of one or more embodied agents.
In this paper we propose a strategy learning model for autonomous agents based on classification. In the literature, the most commonly used learning method in agent-based systems is reinforcement learning. In our opinion, classification can be considered a good alternative. This type of supervised learning can be used to generate a classifier that allows the agent to choose an appropriate action for execution. Experimental results show that this model can be successfully applied for strategy generation even if rewards are delayed. We compare the efficiency of the proposed model and reinforcement learning using the farmer–pest domain and configurations of various complexity. In complex environments, supervised learning can improve the performance of agents much faster that reinforcement learning. If an appropriate knowledge representation is used, the learned knowledge may be analyzed by humans, which allows tracking the learning process.
3
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Purpose: Analysis is made of requirements posed by tasks of agents operating in the intelligent manufacturing systems and their resulting architecture is presented. Design/methodology/approach: Architecture of agent systems for industrial environment is presented, making it possible to generate the particular agents customised for the specific tasks, based on the automatic analysis of its required features. Findings: Extension of cellular automata approach underlying the conventional agent behaviour specification using the Fuzzy Cognitive Maps is presented in conjunction with the neural networks providing learning capability of the agents designed for the various levels of the manufacturing supervisory and execution systems. Adding reaction time specification to FCM makes it possible to analyse and design systems with the required behaviour. Research limitations/implications: Specific features of the designed agent architecture have been tested as separate mechanisms which can be merged into the final comprehensive at a later stage. Originality/value: Agent architecture is proposed for the industrial applications of single agents and their groups that can collaborate to achieve the individual and joint goals specified in reaction to changing environment conditions and into their agendas in XML format. Automatic generation of custom agent reactions models can be carried out based on a set of requirements that may be specified in the if-then rules form.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.