San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and...

39
San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1 , Gerold Huber 2 , Quentin Collier 3 and Gordon K. Lee 1

Transcript of San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and...

Page 1: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies

 

 Christopher Paolini

1, Gerold Huber

2, Quentin Collier

3 and Gordon K. Lee

1

Page 2: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Outline of Presentation:

The Mobile Robotic System Overview

The ANFIS Algorithm

Sensor Integration

Graphics User-Interface

Results

Conclusions and Future Work

Page 3: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Goal:

Develop a mobile robotic testbed to investigate several control algorithms and sensor fusion techniques

Approach: Use web-based video streaming and embedded control architecture for flexibility and robustness

Page 4: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Page 5: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

iRobot Create® Platform

• Single Board Computer (SBC)

• Linux Voyage

• Unibrain Fire-i™ Digital Camera

• Linksys Wireless-G PC adapter card

• iRobot Create® platform

• Single board computer (SBC)

• Voyage Linux (Debian)

• Unibrain Fire-i™ digital camera (IEEE 1394)

• Proxima 802.11g PCMCIA adapter card with external 5db gain antenna

Page 6: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

iRobot Create® with Sensor Arrays

Page 7: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

• iRobot Create® based robot designed with several sensors

• Streaming video

• Web teleoperation

• ANFIS automation

Experimental iRobot developed at San Diego State University

Unibrain Fire-i™ Digital Camera

802.11g PCMCIA

Transceiver

360° Ultrasonic Sensor Array

Arduino Mega MCU

2.5dBi gain indoor omni-

directional antenna

Thermal Sensor

Migrus C787 DCF-P single board

computer with a 1.2GHz Eden ULV

Processor

IR Sensor

9DOF Inertial Measurement Unit

Page 8: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Sensors

3-Axis Rate Gyroscope 3-Axis Accelerometer 3-Axis Magnetometer

Inertial Measurement Unit Controller

Page 9: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Ultrasonic sensor array using an Arduino Mega 2560 microcontroller

Page 10: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

MaxBotix LV-EZ1 Ultrasonic Sensor

An array of 10 MaxBotix LV-EZ1 sensors suspended

on two circular plates.

Each MaxBotix sensor provides a 36 degree FOV.

Ultrasonic Sensor Array

Page 11: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Thermal Sensor

Page 12: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

How the iRobot Adjusts Its Heading

The iRobot changes its azimuth by sending a 16 bit signed

value in the range [-2000, 2000] mm that defines a

turning radius

The turning radius is a ray from the center of the turning

circle to the center of the robot

r > 0 robot turns left

r < 0 robot turns right

Special cases: r = 32768 or 32767 (0x8000 or 0x7FFF)

causes robot to move straight

r = 0xFFFF robot turns in place clockwise

r = 0x0001 robot turns in place counter-clockwise

r2r1

Large radius small curvature

Small radius large curvature

Page 13: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

How the Web GUI Determines the Turning Radius

(x,y)

1tanx

y

23.8 2142.79r

In quadrant I and IV for >5

In quadrant II and III for <-5

23.8 2142.79r III

III IV

Assume r varies linearly with

Page 14: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

• Robot has an onboard 9DOF IMU

• Incorporates four sensors: LY530ALH single-axis gyro, LPR530AL dual-axis gyro, ADXL345 triple-axis accelerometer, and a HMC5843 triple-axis magnetometer

• Gives nine degrees of inertial measurement

• LY530ALH: STMicroelectronics ±300 °/s analog yaw-rate gyroscope

• HMC5843: Honeywell HMC5843, a 3-axis digital magnetometer outputs Euler X,Y,Z orientation vectors and roll () and pitch () angles (tilt sensor) with 12-bit ADC at 10 Hz

• MCU computes azimuth or “yaw” with an accuracy of 1-2 º

Inertial Measurement Unit (IMU)

1

cos sin sin cos sin

cos sin

tan

h

h

h

h

x x y z

y y z

yAz

x

Page 15: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

• We want the absolute bearing defined by the remote user with the GUI to equal the absolute bearing reported by the IMU

• How to compensate for unknown system dynamics?

• Can define a neural network to model robot (plant) dynamics and use training data to tune network parameters

• Define a set of 4-tuple training data:

which are the ith desired (from GUI) azimuth, azimuth rate, actual (from IMU) azimuth and azimuth rate, respectfully.

How to Model Robot Dynamics while Turning?

1 1 1 1

2 2 2 2d d a a

d d a a

n n n nd d a a

id

id

ia

ia

From the LY530ALH yaw rate

sensor

From the HMC5843 digital

magnetometer

Page 16: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

ADAPTIVE AJAX-BASED STREAMING VIDEO SYSTEM

AJAX based Web interface for telerobotic control of the iRobot Create Encoding bit rate as a function of fps

Page 17: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

Virtual Force Field Approach for Obstacle Avoidance

Page 18: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

The Certainty Grid

• 2-Dimensional array of cells

• Each cell contains a Certainty Value (CV)

• CV indicates the measure of confidence that an object exist within a cell

• Instantaneous map for obstacle representation

• dx = dy = 15 cm

• 21 by 21 square cells represent the Certainty Grid

Fron

t

Page 19: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

Method for Updating Certainty Values

Each sensor corresponds to a particular angle Ө, based on

its position on the sensor assembly

At a given time, a sensor returns a distance d

Eq. 1 and 2 transform (d, Ө) → (x’,y’)

(1)

(2)

Page 20: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

Obstacle Avoidance for Path Planning Task

Port S

ide

Front

Page 21: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Cells Located on the Acoustic Axis

Page 22: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

(a) Histogram Grid ; (b) Snapshot of Video Camera

Page 23: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

A1

A2

B1

B2

w 1

w 2

f

x y

x y

x

y

layer 1

layer 2 layer 3

layer 4

layer 5

w 2

w1 w f1 1

w f2 2

If x is Ai and y is Bj, then iiii ryqxpf

i

n

1iifwf

r

The ANFIS Architecture

Page 24: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Forward pass: consequent parameters

X i1 Xi Si1 ai1(b i1T ai1

T Xi )

Si1 Si Sia i1ai1

T Si

1 ai1T Siai1

, i 0, 1, .. ., P 1

E Ei (ydi y i)2 ei

2

i1

P

i1

P

i1

P

Off-line Training

Backwards pass: premise parameters

Page 25: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Forward pass: consequent parameters

On-line Learning

Backwards pass: premise parameters

Page 26: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Page 27: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Simulation Results

Page 28: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

System Integration through an Arduino MCU

ANFIS controller is being implemented on an Arduino Mega that uses the magnetometer and accelerometer output for path tracking.

Initial simulation studies have been conducted using a MISO controller to test this proof of concept.

Then performed actual experimentation using one and two inputs to the ANFIS controller

• Multiple sensors: thermal (person/fixture differentiation), ultrasonic (collision avoidance and path planning), IR (automatic docking), video (telerobotic control), and magnetometer and accelerometer (orientation and position) have been tested and integrated into the overall system architecture.

Page 29: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

Initial Web Browser Interface

Page 30: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Page 31: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Experimental Results

Bearing Scenario

Page 32: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Bearing Scenario

Page 33: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Constant Turn

Page 34: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Following Scenario

Page 35: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Learning Scenario

Page 36: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

ANFIS Responding Time             

       Response Time

 

  Inputs Membership-functions ANFIS Computation Time 

 

          

  1 2 3-4 16-18   

 

          

    3 7-8 20-23   

 

          

    4 9-11 23-26   

 

          

    5 13-15 26-29   

 

          

    6 17-19 30-32   

 

          

    7 22-23 36-38   

 

          

  2 2 12-14 26-28   

 

          

    3 30-31 44-46   

 

          

    4 80-82 94-96   

 

          

Page 37: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Conclusions and Future Work

•A MIMO ANFIS controller has been designed and tested through simulation and experimental studies

•The desired controller can adaptively adjusting to system variations through supervised and un-supervised learning.

•Future tasks include extending the MIMO design to a multiple inputs, two output structure and evaluate the performance of this MIMO implementation using Player/Stage and experimentation.

•We will add on-line learning functionality to our embedded MIMO ANFIS that will effectively tune the parameters computed from off-line training data.

Page 38: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering

Page 39: San Diego State University College of Engineering A Web-Based Mobile Robotic System for Control and Sensor Fusion Studies Christopher Paolini 1, Gerold.

San Diego State University

College of Engineering