Chapter 8. Learning of Gestures by Imitation in a Humanoid Robot in Imitation and Social Learning in Robots, Calinon and Billard. Course: Robots Learning.

Slides:



Advertisements
Similar presentations
Gestures Recognition. Image acquisition Image acquisition at BBC R&D studios in London using eight different viewpoints. Sequence frame-by-frame segmentation.
Advertisements

We consider situations in which the object is unknown the only way of doing pose estimation is then building a map between image measurements (features)
Active Appearance Models
Lecture 16 Hidden Markov Models. HMM Until now we only considered IID data. Some data are of sequential nature, i.e. have correlations have time. Example:
Robust Speech recognition V. Barreaud LORIA. Mismatch Between Training and Testing n mismatch influences scores n causes of mismatch u Speech Variation.
Proposed concepts illustrated well on sets of face images extracted from video: Face texture and surface are smooth, constraining them to a manifold Recognition.
Designing Facial Animation For Speaking Persian Language Hadi Rahimzadeh June 2005.
On-Line Probabilistic Classification with Particle Filters Pedro Højen-Sørensen, Nando de Freitas, and Torgen Fog, Proceedings of the IEEE International.
2004/11/161 A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition LAWRENCE R. RABINER, FELLOW, IEEE Presented by: Chi-Chun.
Hidden Markov Models Theory By Johan Walters (SR 2003)
Foundations of Statistical NLP Chapter 9. Markov Models 한 기 덕한 기 덕.
Hidden Markov Models in NLP
Hidden Markov Model based 2D Shape Classification Ninad Thakoor 1 and Jean Gao 2 1 Electrical Engineering, University of Texas at Arlington, TX-76013,
Bayesian Robust Principal Component Analysis Presenter: Raghu Ranganathan ECE / CMR Tennessee Technological University January 21, 2011 Reading Group (Xinghao.
Hilbert Space Embeddings of Hidden Markov Models Le Song, Byron Boots, Sajid Siddiqi, Geoff Gordon and Alex Smola 1.
HMM-BASED PATTERN DETECTION. Outline  Markov Process  Hidden Markov Models Elements Basic Problems Evaluation Optimization Training Implementation 2-D.
1 Robust Video Stabilization Based on Particle Filter Tracking of Projected Camera Motion (IEEE 2009) Junlan Yang University of Illinois,Chicago.
A Data-Driven Approach to Quantifying Natural Human Motion SIGGRAPH ’ 05 Liu Ren, Alton Patrick, Alexei A. Efros, Jassica K. Hodgins, and James M. Rehg.
Engineering Data Analysis & Modeling Practical Solutions to Practical Problems Dr. James McNames Biomedical Signal Processing Laboratory Electrical & Computer.
Dynamic Time Warping Applications and Derivation
Laurent Itti: CS599 – Computational Architectures in Biological Vision, USC Lecture 7: Coding and Representation 1 Computational Architectures in.
Handwritten Character Recognition using Hidden Markov Models Quantifying the marginal benefit of exploiting correlations between adjacent characters and.
Statistical Natural Language Processing. What is NLP?  Natural Language Processing (NLP), or Computational Linguistics, is concerned with theoretical.
Database Construction for Speech to Lip-readable Animation Conversion Gyorgy Takacs, Attila Tihanyi, Tamas Bardi, Gergo Feldhoffer, Balint Srancsik Peter.
CS Machine Learning. What is Machine Learning? Adapt to / learn from data  To optimize a performance function Can be used to:  Extract knowledge.
Information Retrieval in Practice
Function Approximation for Imitation Learning in Humanoid Robots Rajesh P. N. Rao Dept of Computer Science and Engineering University of Washington,
Isolated-Word Speech Recognition Using Hidden Markov Models
1 Robust HMM classification schemes for speaker recognition using integral decode Marie Roch Florida International University.
© H. Hajimirsadeghi, School of ECE, University of Tehran Conceptual Imitation Learning Based on Functional Effects of Action Hossein Hajimirsadeghi School.
Cut-And-Stitch: Efficient Parallel Learning of Linear Dynamical Systems on SMPs Lei Li Computer Science Department School of Computer Science Carnegie.
Multimodal Interaction Dr. Mike Spann
Segmental Hidden Markov Models with Random Effects for Waveform Modeling Author: Seyoung Kim & Padhraic Smyth Presentor: Lu Ren.
Fundamentals of Hidden Markov Model Mehmet Yunus Dönmez.
STUDY, MODEL & INTERFACE WITH MOTOR CORTEX Presented by - Waseem Khatri.
Bayesian networks Classification, segmentation, time series prediction and more. Website: Twitter:
Cognitive Computer Vision Kingsley Sage and Hilary Buxton Prepared under ECVision Specific Action 8-3
Online Arabic Handwriting Recognition Fadi Biadsy Jihad El-Sana Nizar Habash Abdul-Rahman Daud Done byPresented by KFUPM Information & Computer Science.
Using Inactivity to Detect Unusual behavior Presenter : Siang Wang Advisor : Dr. Yen - Ting Chen Date : Motion and video Computing, WMVC.
ECE 8443 – Pattern Recognition LECTURE 10: HETEROSCEDASTIC LINEAR DISCRIMINANT ANALYSIS AND INDEPENDENT COMPONENT ANALYSIS Objectives: Generalization of.
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: ML and Simple Regression Bias of the ML Estimate Variance of the ML Estimate.
A Passive Approach to Sensor Network Localization Rahul Biswas and Sebastian Thrun International Conference on Intelligent Robots and Systems 2004 Presented.
ECE 8443 – Pattern Recognition LECTURE 08: DIMENSIONALITY, PRINCIPAL COMPONENTS ANALYSIS Objectives: Data Considerations Computational Complexity Overfitting.
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition Objectives: Reestimation Equations Continuous Distributions.
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition Objectives: Reestimation Equations Continuous Distributions.
Real-Time Simultaneous Localization and Mapping with a Single Camera (Mono SLAM) Young Ki Baik Computer Vision Lab. Seoul National University.
CS Statistical Machine learning Lecture 24
CHAPTER 8 DISCRIMINATIVE CLASSIFIERS HIDDEN MARKOV MODELS.
The geometry of the system consisting of the hyperbolic mirror and the CCD camera is shown to the right. The points on the mirror surface can be expressed.
Chapter 7. Learning through Imitation and Exploration: Towards Humanoid Robots that Learn from Humans in Creating Brain-like Intelligence. Course: Robots.
ECE 8443 – Pattern Recognition ECE 8423 – Adaptive Signal Processing Objectives: Supervised Learning Resources: AG: Conditional Maximum Likelihood DP:
Probabilistic reasoning over time Ch. 15, 17. Probabilistic reasoning over time So far, we’ve mostly dealt with episodic environments –Exceptions: games.
From Genomics to Geology: Hidden Markov Models for Seismic Data Analysis Samuel Brown February 5, 2009.
 Present by 陳群元.  Introduction  Previous work  Predicting motion patterns  Spatio-temporal transition distribution  Discerning pedestrians  Experimental.
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition Objectives: Elements of a Discrete Model Evaluation.
CS Statistical Machine learning Lecture 12 Yuan (Alan) Qi Purdue CS Oct
Discriminative Training and Machine Learning Approaches Machine Learning Lab, Dept. of CSIE, NCKU Chih-Pin Liao.
Statistical Models for Automatic Speech Recognition Lukáš Burget.
1 Hidden Markov Model: Overview and Applications in MIR MUMT 611, March 2005 Paul Kolesnik MUMT 611, March 2005 Paul Kolesnik.
CS Statistical Machine learning Lecture 25 Yuan (Alan) Qi Purdue CS Nov
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition Objectives: Reestimation Equations Continuous Distributions.
A Hybrid Model of HMM and RBFN Model of Speech Recognition 길이만, 김수연, 김성호, 원윤정, 윤아림 한국과학기술원 응용수학전공.
Definition of the Hidden Markov Model A Seminar Speech Recognition presentation A Seminar Speech Recognition presentation October 24 th 2002 Pieter Bas.
Statistical Models for Automatic Speech Recognition
Outline S. C. Zhu, X. Liu, and Y. Wu, “Exploring Texture Ensembles by Efficient Markov Chain Monte Carlo”, IEEE Transactions On Pattern Analysis And Machine.
Statistical Models for Automatic Speech Recognition
Dimension reduction : PCA and Clustering
LECTURE 15: REESTIMATION, EM AND MIXTURES
Speech recognition, machine learning
Speech recognition, machine learning
Presentation transcript:

Chapter 8. Learning of Gestures by Imitation in a Humanoid Robot in Imitation and Social Learning in Robots, Calinon and Billard. Course: Robots Learning from Humans Ryun, Seokyun Human Brain Function Laboratory Interdisciplinary Program in Neuroscience Seoul National University

Contents Introduction Experimental Set-up Data Processing –Preprocessing by principal component analysis (PCA) –Encoding in hidden Markov models (HMMs) –Recognition, Data retrieval, Imitation metrics Results and Performance of the System Discussion of the Model Conclusion Discussion (further) 2

Introduction Imitation abilities in human ?

Introduction Robot learning by imitation (robot programming by demonstration) explores novel means of implicitly teaching a robot new motor skills Advantages -Provide a natural, user-friendly means of implicitly programming the robot -Constrain the search space of motor learning by showing possible and/or optimal solutions Gesture recognition and reproduction Writing letters, arm trajectory… How to find a representation of the data that encapsulates only the key aspects of the gesture…?

Experimental Set-up

Training: by eight healthy volunteers (imitating six motions in video) Knocking on a door Raising a glass, drinking and putting it back on a table Waving goodbye Drawing a stylized alphabet letters A, B and C Data recording: by three x-sens motion sensors (location, joint angle) Sampling rate: 100 Hz by Color-based vision system (camera): arm tracking by markers

Data Processing

Preprocessing by principal component analysis (PCA) Determining the directions along which the variability of the data is maximal Minimize the statistical dependence across the data Dimensionality reduction Advantages Reduce noise Reduce dimensionality (faster processing) Simple representation

Data Processing Hidden Markov Models (HMMs) X = states, y = possible observations, a = transition probabilities, b = output probabilities Robust statistical algorithm in time-dependent sequential events (speech recognition, gesture recognition, etc.)

Data Processing Encoding in hidden Markov models (HMMs) A set of time series is used to train a HMM with I+F output variables. The parameters are expressed as a set of parameters (initial state distribution, state transition probabilities, means of output variables, standard deviation of the output variables, respectively)

Data Processing Training Transition probabilities and the observation distributions are estimated by expectation maximization (EM) algorithm Optimal number of state selection : Bayesian information criterion (BIC) L = likeligood of the fitted model n(p) = number of independent parameters in HMM T = number of observation data used in fitting the model

Data Processing Recognition Estimate the likelihood that the new signals could have been generated by one of the models Data retrieval Sequence of state is reconstructed by the Viterbi algorithm HMM states  location/angle sequences

Data Processing Imitation metrics Determining a metric of imitation performance Error estimation: by cost function In univariate case e = mean error of the observed data D’ compared to the dataset D u(t) = sequence of means associated with the sequence of states e’ = RMS error, x(t) = demonstration datasets, x’(t) = reproduction datasets

Data Processing Error estimation: In multidimensional case K = total dimensions, w = weights vector (importance of variable i) Sigma = standard deviation of each time-point/variable

Data Processing

Results and Performance of the System Recognition rate = 96 % (23 motions correct)

Results and Performance of the System

Discussion of the Model PCA: Making the HMM encoding more robust The advantage of encoding the signals in HMMs Better generalization of the data Robust to distortion in time (e.g. drinking timing: fast or slow) It would be interesting to extend the model to using asynchronous HMMs Joint probability of pairs of asynchronous sequences describing the same sequence of events (e.g. visual lip reading and audio signals) Visual + motor representation  alignment

Discussion of the Model Similarity with work in psychology and ethology Imitation using string parsing

Discussion of the Model Similarity with work in psychology and ethology Associative sequence learning (ASL)

Discussion of the Model Similarity with work in psychology and ethology Algebraic framework for the correspondence problem

Conclusion Implementation of an HMM-based system to encode, generalize, recognize and reproduce gestures, with representation of the data in visual and motor coordinates Link between theoretical concepts and practical applications Observed elements of a demonstration, and the organization of these elements, should be stochastically described to have a robust robot application

Discussion (further) Mimicking innate sensorimotor representation of human Visual markers  ?  corresponding body parts Self-feedback during imitation (tactile, proprioception)  Exercise HMM approach in imitation: similar to the function of cerebellum HMM in HMM subsets: for parallel and serial sequences Limitation: individualizing the gesture (based on numerous sensory feedback) (e.g. vestibular function, proprioceptive feedback, tactile feedback, movement tracking of own body parts by visual, sensory feedback)

Thank You