This paper presents a method for the biped dynamic walking and balance control using reinforcement learning, which learns without <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a priori</i> knowledge about model. The learning architecture developed is aimed to solve complex problems in robotic actuation by mapping action space from discretized domain continuous one. It employs discrete actions construct policy action. allows scaling of...

This paper presents an inverse kinematics algorithm for a redundant manipulator. The proposed approach combines analytical and numerical methods to solve generalized problems simultaneously in both free constrained working spaces. method introduces virtual repulsive torque imposed by the topological configuration of robot links, joint limits, obstacles with motion trajectories. calculates shortest distances between limit, obstacles. Links manipulator are modeled ellipsoids so as evaluate...

Techniques for transferring human behaviors to robots through learning by imitation/demonstration have been the subject of much study. However, direct transfer motion trajectories humanoid does not result in dynamically stable robot movements because differences and kinematics dynamics. An imitating algorithm called posture-based imitation with balance (Post-BL) is proposed this paper. This Post-BL consists three parts: a key posture identification method used capture postures as knots...

In a multiagent system, if agents' experiences could be accessible and assessed between peers for environmental modeling, they can alleviate the burden of exploration unvisited states or unseen situations so as to accelerate learning process. Since how build up an effective accurate model within limited time is important issue, especially complex environments, this paper introduces model-based reinforcement method based on tree structure achieve efficient modeling less memory consumption....

A Dyna-Q algorithm is known as model-based reinforcement learning, so the learning agent not only interacts with environment to learn an optimal policy, but also builds environmental model simultaneously. To deal shortage of online samples, introduced achieve goal. enhance efficiency model, this paper proposes a shaping method compensate for bleak states scarcely visited during neighbor information. After acquiring accurate many virtual experiences are sampled from and indirect thereby...

This study is about the manufacturing of a personified automatic robotic lawn mower with image recognition. The system structure that platform above crawler tracks combined mower, steering motor, slide rail, and webcam to achieve purpose personification. Crawler strong grip good ability adapt terrain are selected as moving vehicle simulate human feet. In addition, mechanism designed left right swing mowing promote efficiency innovation, then eyes replaced by Webcam identify obstacles. A...

This correspondence presents a multistrategy decision making system for robot soccer games. Through reinforcement processes, the coordination between robots is learned in course of game. Meanwhile, better action can be granted after an iterative learning process. The experimental scenario five-versus-five game, where proposed dynamically assigns each player to position primitive role, such as attacker, goalkeeper, etc. responsibility varies along with change role state transitions....

This study introduces a method to enable robot learn how perform new tasks through human demonstration and independent practice. The proposed process consists of two interconnected phases; in the first phase, state-action data are obtained from demonstrations, an aggregated state space is learned terms decision tree that groups similar states together reinforcement learning. Without postprocess trimming, induction, encodes control policy can be used by means repeatedly improving itself. Once...

A collision-free trajectory planning method based on speed alternation strategy for multijoint manipulators in overlapped working envelopes is proposed. Since the shape of a robot's link usually rectangular or cylindrical approximately, proposed models mathematically by quadric primitives, such as ellipsoids and spheres. The occurrence collisions between links can be predicted easily means relative coordinate transformations geometric deformations those ellipsoids. Furthermore,...

In this paper, an Fuzzy Markup Language (FML)-based type-2 fuzzy ontology is proposed to represent the computer Go knowledge, including FML transformation mechanism, a set construction, and inference mechanism. The mechanism transforms generated smart game format (SGF) files into FML-based document describe ontology. construction responsible for building sets. Based on built sets, infers possibility of game's winning rate. It hoped that idea are feasible inferring rate games in future.

A model-based reinforcement learning (RL) method which interplays direct and indirect to update <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${Q}$ </tex-math></inline-formula> functions is proposed. The environment approximated by a virtual model that can predict the transition next state reward of domain. This used train accelerate policy learning. Lookup table methods are usually establish such...

Game of Go is one the main challenges in artificial intelligence. In particular, it much harder than chess, spite fact that fully observable and has very intuitive rules. Computer been developing for past several years, NoGo game similar to Go, sense each player puts a stone on board alternatively, stones do not move. However, goal different, example, first who either suicides or kills group lost game. this paper, Fuzzy Markup Language (FML) applied infer position good move new NoGo. The...

State value estimating is an important issue in reinforcement learning. It affects the performance significantly. The methods of lookup tables have advantages convergence rate. But they need prior knowledge about how to partition state space advance. also not reasonable a real system since values associated with different sensory inputs but belonging representing are same. We proposed method discretize adaptively and effectively terms approach akin decision tree methods. In each...

Q-learning, a most widely used reinforcement learning method, normally needs well-defined quantized state and action spaces to obtain an optimal policy for accomplishing given task. This means it difficult be applied real robot tasks because of poor performance learned behavior due the failure quantization continuous spaces. In this paper, we proposed fuzzy-based CMAC method calculate contribution values estimate value in order make motion smooth effective. And implement multi-agent system...

This article presented a Dyna-Q learning method based on world model of tree structures to enhance the efficiency sampling data in reinforcement problem. The Q-Learning mechanism is for policy as by observing transitions between states after actions taken. In early stages learning, agent does not have an accurate but explores environment possible collect sufficient experiences approximate model. When develops more model, planning can use produce simulated accelerate value iterations. Thus,...

Abstract Dyna-Q, a well-known model-based reinforcement learning (RL) method, interplays offline simulations and action executions to update Q functions. It creates world model that predicts the feature values in next state reward function of domain directly from data uses train functions accelerate policy learning. In general, tabular methods are always used Dyna-Q establish model, but needs many more samples experience approximate environment concisely. this article, an adaptive method...

In this paper, a model learning method based on tree structures is present to achieve the sample efficiency in stochastic environment. The proposed composed of Q-Learning algorithm form Dyna agent that can used speed up learning. learn policy, and for builds environment simulates virtual experience. experience decrease interaction between make perform value iterations quickly. Thus, has additional updating policy. simulation task, mobile robot maze, introduced compare methods, Q-Learning,...

In this study, a tour recommendation system based on social media photos is proposed. The proposed can generate trip tours considering both the user’s current location and interests. First, we exploited geotagged photo dataset from websites, which includes related information such as user ID numbers, timestamps, hashtags, GPS coordinates. With information, second step to group identify those places that could be considered relevant for travellers using clustering algorithms. third...

<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$ </tex-math></inline-formula> -learning is a generic approach that uses finite discrete state and an action domain to estimate values using tabular or function approximation methods. An intelligent agent eventually learns policies from continuous sensory inputs encodes these environmental onto space. The application of in state/action the subject many...