Robot Intelligence - An Advanced Knowledge Processing Approach

Preface

Contents

Contributors

Programming-by-Demonstration of Robot Motions

Introduction

Learning from Human Demonstration

Interpretation of Demonstrations in Hand-State Space

Skill Encoding Using Fuzzy Modeling

Generation and Execution of Robotic Trajectories Based on Human Demonstration

Mapping Between Human and Robot Hand States

Definition of Hand-States for Specific Robot Hands

Next-State-Planners for Trajectory Generation

Demonstrations of Pick-and-Place Tasks

Variance from Multiple Demonstrations

Experimental Platform

Experimental Evaluation

Experiment 1: Learning from Demonstration

Importance of the Demonstration

Experiment 2: Generalization in Workspace

Experiment 3: a Complete Pick-and-Place Task

Conclusions and Future Work

References

Grasp Recognition by Fuzzy Modeling and Hidden Markov Models

Introduction

An Experimental Platform for PBD

Simulation of Grasp Primitives

Geometrical Modeling

Modeling of Inverse Kinematics

Modeling of Grasp Primitives

Modeling by Time-Clustering

Training of Time Cluster Models Using New Data

Recognition of Grasps-Three Methods

Recognition of Grasps Using the Distance Between Fuzzy Clusters

Recognition Based on Qualitative Fuzzy Recognition Rules

Distance Norms

Extrema in the Distance Norms and Segmentation

Set of Fuzzy Rules

Similarity Degrees

Recognition Based on Time-Cluster Models and HMM

Experiments and Simulations

Time Clustering and Modeling

Grasp Segmentation and Recognition

Conclusions

References

Distributed Adaptive Coordinated Control of Multi-Manipulator Systems Using Neural Networks

Introduction

Preliminaries

Multi-Manipulator System Description

Radial Basis Function Neural Network

Controller Design

Performance Analysis

Simulation Example

Conclusion

References

A New Framework for View-Invariant Human Action Recognition

Introduction

Overview of the Proposed Approach

Exemplar Selection and Representation

Key Pose Extraction

2D Silhouette Image Generation

Contour Shape Feature

Action Modelling and Recognition

Exemplar-based Hidden Markov Model

Action Modelling

Action Recognition

Action Category Revalidation

Experiments

Conclusion

References

Using Fuzzy Gaussian Inference and Genetic Programming to Classify 3D Human Motions

Introduction

Human Skeletal Representation

The Learning Method

Model Description: the Fuzzy Membership Function

Model Generation: Fuzzy Gaussian Inference

Membership Evaluation

Mathematical Properties

Extracting Fuzzy Rules Using Genetic Programming

Experiment and Results

Apparatus

Participants

Procedure

Results

Discussion

Conclusion

References

Obstacle Detection Using Cross-Ratio and Disparity Velocity

Introduction

Background

Algorithm Overview

Generation of Mesh Maps

Mesh Generation

Estimation of the Ground Floor

Identification of Safe Regions within the Ground Plane

Incremental Addition of Feature Points

Safe Path Detection

Further Evaluation

Estimation of Ground Floor

Obstacle Detection

Summary

References

Learning and Vision-Based Obstacle Avoidance and Navigation

Introduction

Depth Perception

Absolute Depth and Binocular Vision

Absolute Depth

Relative Depth and Monocular Vision

Edge Direction and Perspective

Clarity of Detail and Texture Gradient

Size Cues

Motion Depth Cues

Occlusion

Why Learning and How to Learn for Monocular Visions

The Role of Experience

Learning Methods

MILN Learning

Special Problem: Illumination Changes in Outdoor Scenes

Finding Passable Regions for Obstacle Avoidance from Single Image Using MILN

Feature Vector

Edge

Clarity of Detail and Texture Gradient

Color Similarity with Lighting Invariance

Pixel Position and Region Connection

Training Data Generation and Experiment

Performance Evaluation

Control Law and Navigation

From Obstacle Boundaries to Motor Commands

Discussion

Learning Ability

Changing Lighting Conditions

Learning from Experience

References

A Fraction Distortion Model for Accurate Camera Calibration and Correction

Introduction

Previous Work

The Proposed Work

A New Distortion Model

A Novel Calibration Algorithm

Pin-Hole Camera Model

Optimisation of All Parameters

The Correction of the Distorted Image Points

Summary of the Novel Camera Calibration and Correction Algorithm

Experimental Results

Synthetic Data

Calibration and Correction

Collinearity Constraint

Different Levels of Noise

Real Images

Conclusion

References

A Leader-Follower Flocking System Based on Estimated Flocking Center

Introduction

Flocking System

Flocking Algorithms

Algorithm Stability

Experiments

Simulations

Conclusions

References

A Behavior Based Control System for Surveillance UAVs

Introduction

Platform and Atomic Actions

UAV Platform

System Structure

Atomic Actions

Software Architecture and Behavior Development

Software Architecture

Behavior Development

Ground Behavior

Takeoff Behavior

Hovering Behavior

GPS Landing Behavior

Vision Landing Behavior

Emergency Landing Behavior

GPS Tracking Behavior

Vision Tracking Behavior

Obstacle Avoidance Behavior

Trajectory Tracking Behavior

Vision Module Development

SURF Algorithm

Coordination Transformation

Kalman Filter

Experiment Results

Hovering Behavior

Vision Tracking Behavior

Trajectory Tracking Behavior

Trajectory Tracking Behavior with Obstacle Avoidance capability

GPS Landing Behavior

Vision Landing Behavior

Conclusion and Future Work

References

Hierarchical Composite Anti-Disturbance Control for Robotic Systems Using Robust Disturbance Observer

Introduction

Formulation of the Problem

Hierarchical Composite Anti-Disturbance Control (HCADC)

Applications to a Two-Link Robotic System

Conclusions

Proof of the Lemma 11.1

Proof of the Lemma 11.2

References

Autonomous Navigation for Mobile Robots with Human-Robot Interaction

Introduction

Human-Robot Interaction

Subject Following with Target Pursuing

Correspondence

Multi-Cue Integration

Robust Tracking

Pursuing

Mapping

Qualitative Localization

Scene Association

Scene Recognition

Planning and Navigation

Conclusion

References

Prediction-Based Perceptual System of a Partner Robot for Natural Communication

Introduction

Prediction-Based Perceptual System for A Partner Robot

A Partner Robot; Hubot

A Prediction-Based Perceptual System

Perceptual Modules

Differential Extraction

Human Detection

Object Detection

Hand Motion Recognition

Architecture of Prediction Based Perceptual System

Input Layer Based on Spiking Neurons

Clustering Layer Based on Unsupervised Learning

Prediction Layer and Perceptual Module Selection Layer

Update of Learning Rate for Perceptual Module Selection

Learning for Prediction and Perceptual Module Selection

Experimental Results

Clustering for Prediction

Real-Time Learning in Interaction

Additional Learning

Conclusions

References

Index