1 Assistive Human-Machine Interfaces via Artificial Neural Networks Wei Tech Ang & Cameron N. Riviere The Robotics Institute Carnegie Mellon University.

Slides:



Advertisements
Similar presentations
Lecture 20 Dimitar Stefanov. Microprocessor control of Powered Wheelchairs Flexible control; speed synchronization of both driving wheels, flexible control.
Advertisements

Ch. Eick: More on Machine Learning & Neural Networks Different Forms of Learning: –Learning agent receives feedback with respect to its actions (e.g. using.
CSC321: 2011 Introduction to Neural Networks and Machine Learning Lecture 7: Learning in recurrent networks Geoffrey Hinton.
Channel Estimation for Mobile OFDM
Classification Neural Networks 1
Inferring Hand Motion from Multi-Cell Recordings in Motor Cortex using a Kalman Filter Wei Wu*, Michael Black †, Yun Gao*, Elie Bienenstock* §, Mijail.
Perceptron.
Machine Learning Neural Networks
Introduction to Mobile Robotics Bayes Filter Implementations Gaussian filters.
SA-1 Body Scheme Learning Through Self-Perception Jürgen Sturm, Christian Plagemann, Wolfram Burgard.
Presentation 1: Noise canceling in 1-D data Seri Rahayu Abd Rauf Fatima Boujarwah Juan Chen Liyana Mohd Sharipp Arti Thumar M2.
November 2, 2010Neural Networks Lecture 14: Radial Basis Functions 1 Cascade Correlation Weights to each new hidden node are trained to maximize the covariance.
Discriminative Training of Kalman Filters P. Abbeel, A. Coates, M
Goals of Adaptive Signal Processing Design algorithms that learn from training data Algorithms must have good properties: attain good solutions, simple.
Back-Propagation Algorithm
Artificial Neural Networks
Face Processing System Presented by: Harvest Jang Group meeting Fall 2002.
Neural Networks. Background - Neural Networks can be : Biological - Biological models Artificial - Artificial models - Desire to produce artificial systems.
Kalman filter and SLAM problem
Case Studies Dr Lee Nung Kion Faculty of Cognitive Sciences and Human Development UNIVERSITI MALAYSIA SARAWAK.
West Virginia University
„Bandwidth Extension of Speech Signals“ 2nd Workshop on Wideband Speech Quality in Terminals and Networks: Assessment and Prediction 22nd and 23rd June.
Knowledge Systems Lab JN 9/10/2002 Computer Vision: Gesture Recognition from Images Joshua R. New Knowledge Systems Laboratory Jacksonville State University.
Presentation on Neural Networks.. Basics Of Neural Networks Neural networks refers to a connectionist model that simulates the biophysical information.
Leslie Luyt Supervisor: Dr. Karen Bradshaw 2 November 2009.
Artificial Neural Networks
ARTIFICIAL NEURAL NETWORKS. Introduction to Neural Networks.
1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 23 Nov 2, 2005 Nanjing University of Science & Technology.
Machine Learning Chapter 4. Artificial Neural Networks
STUDY, MODEL & INTERFACE WITH MOTOR CORTEX Presented by - Waseem Khatri.
1 Machine Learning The Perceptron. 2 Heuristic Search Knowledge Based Systems (KBS) Genetic Algorithms (GAs)
NEURAL NETWORKS FOR DATA MINING
Artificial Neural Networks. The Brain How do brains work? How do human brains differ from that of other animals? Can we base models of artificial intelligence.
Robust localization algorithms for an autonomous campus tour guide Richard Thrapp Christian Westbrook Devika Subramanian Rice University Presented at ICRA.
1 Introduction to Neural Networks And Their Applications.
Operant Conditioning of Cortical Activity E Fetz, 1969.
Statistical learning and optimal control: A framework for biological learning and motor control Lecture 4: Stochastic optimal control Reza Shadmehr Johns.
Motor Control. Beyond babbling Three problems with motor babbling: –Random exploration is slow –Error-based learning algorithms are faster but error signals.
Akram Bitar and Larry Manevitz Department of Computer Science
J.-Y. Yang, J.-S. Wang and Y.-P. Chena, Using acceleration measurements for activity recognition: An effective learning algorithm for constructing neural.
Review: Neural Network Control of Robot Manipulators; Frank L. Lewis; 1996.
Neural Network Basics Anns are analytical systems that address problems whose solutions have not been explicitly formulated Structure in which multiple.
Lecture 5 Neural Control
Brain-Machine Interface (BMI) System Identification Siddharth Dangi and Suraj Gowda BMIs decode neural activity into control signals for prosthetic limbs.
C - IT Acumens. COMIT Acumens. COM. To demonstrate the use of Neural Networks in the field of Character and Pattern Recognition by simulating a neural.
CHEE825 Fall 2005J. McLellan1 Nonlinear Empirical Models.
Bab 5 Classification: Alternative Techniques Part 4 Artificial Neural Networks Based Classifer.
Neural Networks Lecture 11: Learning in recurrent networks Geoffrey Hinton.
Robot Intelligence Technology Lab. 10. Complex Hardware Morphologies: Walking Machines Presented by In-Won Park
A Presentation on Adaptive Neuro-Fuzzy Inference System using Particle Swarm Optimization and it’s Application By Sumanta Kundu (En.R.No.
One-layer neural networks Approximation problems
第 3 章 神经网络.
Electric Circuits Chapter 7 Network Frequency Characteristics
Soft Computing Applied to Finite Element Tasks
Intelligent Information System Lab
"Playing Atari with deep reinforcement learning."
Machine Learning Today: Reading: Maria Florina Balcan
General Aspects of Learning
Artificial Neural Networks
Artificial Intelligence Chapter 3 Neural Networks
network of simple neuron-like computing elements
Emre O. Neftci  iScience  Volume 5, Pages (July 2018) DOI: /j.isci
Neural Networks Geoff Hulten.
Artificial Intelligence Chapter 3 Neural Networks
Artificial Intelligence Chapter 3 Neural Networks
Artificial Intelligence Chapter 3 Neural Networks
COSC 4335: Part2: Other Classification Techniques
Artificial Neural Networks / Spring 2002
Akram Bitar and Larry Manevitz Department of Computer Science
Derivatives and Gradients
Presentation transcript:

1 Assistive Human-Machine Interfaces via Artificial Neural Networks Wei Tech Ang & Cameron N. Riviere The Robotics Institute Carnegie Mellon University Pittsburgh, PA 15213, USA Funded by: National Institute of Health Pittsburgh Foundation

2 Introduction  Objective  To create assistive human-machine interfaces to enhance positioning accuracy for patients with movement disorders  Applications  Computer mouse/joystick control  Powered wheel chair control

3 Movement Disorders  Common Types:  Pathological Tremor  any involuntary, approximately rhythmic, and roughly sinusoidal movement  Higher frequency band than voluntary motion  Myoclonic Jerk  sudden muscle contractions that can occur alone or in a sequence  Aperiodic, erratic, unpredictable  Can overlap in frequency with voluntary motion.  Sources:  Multiple Sclerosis (MS), Parkinson diseases, essential tremors etc.

4 Error Compensation Approaches  Tremor Modeling and Compensation  Frequency selective approaches  Low-pass filtering (Riley & Rosen ‘87)  Signal equalizer technique (Gonzalez et al. ‘95)  Adaptive noise canceller (Riviere et al.‘98)  Others: Viscous damping (Beringhause et al. ‘89, Rosen et al. ’95)  Non-Tremulous Error Compensation  Interfaces of sufficiently low bandwidth and input gain  keyboards with large key-pitch or/and ‘sticky operation’  Use other body part for control  Artificial Neural Network approach  simultaneously modeling and canceling both tremulous and non-tremulous types of movement disorder

5  Dynamically adjusted network architecture  Flexibility  Node decoupled extended Kalman filtering learning rule  Faster convergence over backprop Cascade Correlation Neural Networks with Kalman Filtering Output Hidden Input

6 Experimental Data  Collected from 11 test subjects with Multiple Sclerosis (MS) by Univ. of Pittsburgh  Subjects used HeadMaster Plus TM computer head control system (Prentke Romich Company, Wooster, OH)  Icon selection exercise  Move cursor from center of 14” screen (1024 x 768) to a series of circular targets(30-pixel radius)  Dwell in target for > 500ms

7 Neural Networks Training  Multiple Sclerosis (MS)  A serious progressive disease of the central nervous system, caused by malfunction in the immune system  Intention tremor  decent start, tremulous to chaotic trajectory close to target  Training targets  Phase corrected, low- passed trajectory Low-passed trajectory Raw trajectory  Start position □ End position * Target

8 Neural Networks Training  Lack of training data  Screen segments into 8 bearing sectors  N, NE, E, SE, S, SW, W, NW  Each sector we train 2 neural networks  X- & Y-direction  15 input nodes – time series of 15 data points  1 output node – compensated position of 15 th data point   10 hidden nodes N NE SE SW SE W E S 2D cursor trajectory Low-pass & phase correction Forward difference X netY net Training X target Y target X input Y input Split X & Y trajectories

9 Bearing Determination  Exploiting movement disorder characteristics of MS patient:  Intention tremor – decent start, tremulous to chaotic finish  Bearing estimation based on gradients of 1 st 14 data points using maximum likelihood criterion  81.2% success rate  Real-time interactions issues

10 Result - West  Total tests = 29  W – 9  NE – 12  S – 8  Smoother trajectories  Reached and dwelled in target 31.8% (ave) faster Target circle  NN output Raw trajectory  Start position □ End position * Target circle 10-pixel circle

11 Result - Northeast Target circle  NN output Raw trajectory  Start position □ End position * Target circle 10-pixel circle

12 Result - South Target circle  NN output Raw trajectory  Start position □ End position * Target circle 10-pixel circle

13 Result – Decent Trajectories  No over correction  18.4% faster completion time over 4 tests  NN output Raw trajectory  Start position □ End position * Target circle 10-pixel circle

14 Discussion  Demonstrated the feasibility of the ANN approach in modeling and canceling of movement disorders at assistive human- machine interface  The current experiment has not fully exploited the non-linear capability of ANN  Handicapped by the data we inherited  Strength of ANN will become apparent in more complex task scenarios and movement disorders

15 Future Works  Design our own data collection exercise for hand movement disorders  Implement real-time error compensation system  Evaluate subjects’ interaction with system  Extend the method to other type of diseases other than MS, e.g. Parkinson diseases