Hand-Eye Tracker NN Sim Exp Ass

Embed Size (px)

Citation preview

  • 7/31/2019 Hand-Eye Tracker NN Sim Exp Ass

    1/2

    SYS735 Intelligent Control Systems KaCC

    2 Hand-Eye Tracker NN Sim Exp Ass 1 13-Jan-01

    SIMULATION EXPERIMENT #NN1

    Application of Neural Network to Emulate Your Human Eye-Hand Coordination

    for Tracking a Target

    Issued: 23 Jan 06 Due: 2 Feb 06

    Goal. The goal is to train a FF ANN that duplicates your own eye-hand action in tracking a target. You will conduct

    tracking experiments using a PC-based Matlab environment that simulates the dynamics of a motorized camera

    which reveals an IR image of a target on its monitor as shown in Figure 1. Then you will train an ANN to replace

    you as the tracker, as shown in Figure 2.

    Premise. Thoughwe are able to track the target but cant describe what exactly goes on in our thought while we

    were tracking with our eye-hand coordination. So we will let an ANN emulate the tracking pattern.

    Basic programs. The basic Matlab/Simulink programs for this lab are found at www.oakland.edu/~cheok.

    PC-based Simulation/Animation

    Human-in-the-loopEyes

    Decision

    Hand

    Visual Animation

    Target tracking motion

    Dynamics System Simulation

    Dynamics of motorized IR camera

    You & your neurons

    Input Device

    Mouse

    Output Device

    Monitor

    Figure 1. Eye-Hand Tracking Coordination Simulation Experiment

    Visual Animation

    Target tracking motion

    Dynamics System Simulation

    Dynamics of motorized IR camera

    FF Artificial Neural Network

    Emulation of Tracking Skill

    Figure 2. ANN Tracking Coordination Simulation Experiment

    xu

    u

    =

    u

    x

    =

    x

    y

    u

    u

    =

    u

    x

    y

    =

  • 7/31/2019 Hand-Eye Tracker NN Sim Exp Ass

    2/2

    SYS735 Intelligent Control Systems KaCC

    2 Hand-Eye Tracker NN Sim Exp Ass 2 13-Jan-01

    Task 1. Collect Tracking Pattern. You would run the simulation (Figure 1) as explained in class lectures and

    collect the input-output pattern of the eye-hand coordination. Set up the characteristics of the target movement such

    that it moves slowly over the screen (so you can track it). Conduct tracking experiments and collect the data & u .

    Task 2. Proportional-type Controller Premise. In this premise, we assume that the tracking control action is

    somewhat proportional to the errors between the target and camera (crosshairs). So set out to train the ANN inFigure 3 with the error data as inputs. Report on the ANN structure and its trained parameters.

    Task 3. Run the tracking simulation with the P-type trained ANN as shown in Figure 2. Capture the performance

    and comment on what you see and your expectation.

    Task 4. Next increased the speed of target motion, repeat tracking experiments with the P-type ANN. Capture the

    performance and comment on what you see and your expectation. Explain the change in performance you observe.

    Task 5. Proportional & Derivative-type Controller Premise. In this premise, we assume that the tracking control

    action is somewhat proportional to the errors between the target and camera (crosshairs), and also to the rate of the

    target-camera errors. So set out to train the ANN in Figure 3 with the error data and error differences as inputs.

    Report on the ANN structure and its trained parameters.

    Task 6. Run the tracking simulation with the PD-type trained ANN as shown in Figure 2. Capture the performance

    and comment on what you see and your expectation. Explain the change in performance you observe.

    Task 7. Next increased the speed of target motion, repeat tracking experiments with the P-type ANN. Capture the

    performance and comment on what you see and your expectation. Explain the change in performance you observe.

    Task 8. Submit a report with details for the tasks. Provide comments and insights.

    Suggest improvements

    More exciting target motion

    Longer delay for the D in PD

    Model the delay factor in eye-mind-hand

    Artificial Neural Networks

    ( )

    ( )

    2 2 2 2 1 2

    1 1 1 1 1 1

    s s= = +

    = = +

    u f W y b

    y f s s W b

    x

    y

    =

    x

    y

    u

    u

    =

    u

    Figure 3. FF ANN for Proportional-type Controller

    Artificial Neural Networks

    ( )

    ( )

    2 2 2 2 1 2

    1 1 1 1 1 1

    s s= = +

    = = +

    u f W y b

    y f s s W b

    x

    =

    xu

    u

    =

    u

    ( ) ( 1)

    ( ) ( 1)x x

    y y

    k k

    k k

    =

    Figure 3. FF ANN for Proportional & Derivative -type Controller

    The purpose of computing [simulation] is INSIGHT, not [just] numbersHamming