Automatic Pouring Robot

Akilah Harris-WilliamsAdam Olmstead

Philip Pratt-SzeligaWill Roantree

OverviewObjective and MotivationMechanical SystemModeling, Simulation and Verification

TiltPan

Pouring Trajectory GenerationControl Design

TiltPan

Webcam Image ProcessingUncertainty Analysis

Objective and Motivation

To develop an Automatic Pouring RobotIndustrial Applications

Casting RobotHazardous Materials Robot

Purpose

Smoothly pour water from a bottle into a cup without spillingConsists of a bottle, a pan/tilt mechanism, a computer controller, a cup and a webcamThe bottle will be mounted on the mechanismComputer controller will rotate the bottle towards the cupWebcam will view the height of water in the cup.

Mechanical System

Tilt ModelingMass Properties change with tilt angle and water volume. System dynamics are variable.Visual Basic program extracts Center of Mass, Inertial Tensor and Mass from Solidworks API at varying angles and volumes.

Tilt Friction/Gravity Model

Upper Level SystemTilt DynamicsEquation of Motion

Tilt Friction/Gravity ResultsExperimental system response versus Simulated system response to force of gravity.

Tilt Friction/Gravity Cancellation

Pan Modeling

Vaaa 321 )sgn( =++ θθθ

Used full parameter identification methodChirp inputa1=1.9080a2=6.0656a3=8.44290.08 radSteady-State Error

Basic Trajectory

Basic Trajectories

Polynomial Trajectories

Basic Return Trajectories

Return Trajectories

Tilt Control

Needs to follow a smooth trajectorySpecifications:

Mo ≤ 0%, tr ≤ 0.75 s, ess ≤ 1%Used PID controller

System behaved similarly to modelLinear model assumes constant volumeDue to modeling error, needed an easily tweaked controller

Tilt Controller – SimulatedSimulatedMo = 2.87%tr=0.327 sess=0.81%ActualMo = -0.13%tr=0.104 sess=0.138%

Tilt Controller - Actual

Simulation was run assuming a constant volumeValues will change due to:

Constantly changing amount of liquidAny modeling error

Values tweaked from simulation to dampen overshoot

Tilt Controller on Actual Trajectory

Pan ControlUsed for Disturbance Rejection

Needs to have a quick rise time and low steady-state error

SpecificationsMo ≤ 5%, tr ≤ 0.4 s, ess ≤ 2%

Lead-Lag Controller

)9.0)(486.19()266.1)(493.6(30)(

++++

=ss

sssD

Pan Controller - Design

Pan Controller - SimulationSimulated:Mo = 0.367 %tr=0.336 sess=0.037 %Actual:Mo = 5.23 %tr= 0.133 sess= 0.936 %

Tweaked Pan Controller

Actual:Mo=1.55 %tr=0.134sess=0.016 %ts=0.356s (2%)ts=1.548s (1%)

Sensor Subsystem - Webcam

Objective: Pour a specified amount of liquid into a cupNeed to know when to stop pouring to get the desired amountOur solution: Webcam

Image Processing AlgorithmWant to find the volume of liquid in the cup

Find highest pixel row occupied by the colored liquid (determined by green and blue thresholds)Correlate these pixels to actual volumes via experimentation

Webcam Timing

Average frame rate with processing is 13 fps.When the volume in the cup approaches a certain percentage of the desired volume, MATLAB triggers xPC target to stop pouring.Percentage at which the trigger is sent is a function of:

Return trajectoryCommunication delay between MATLAB and xPCTargetFlow rate before stopping

Webcam Data of Sample Pour

Stop Triggered

Uncertainty Analysis

Modeling errorTilt: Friction IdentificationPan: Steady-State Error

Webcam errorNoisy dataSlow feedback

Other sources of errorOccasional encoder spikes

Modeling Error – Tilt Friction ID

Velocities never converge to a steady-state value, due to water sloshing in the bottleDifferent values in each direction

Average was taken

Modeling Error - PanIdentified parameters work fairly well with data used to generate them.

ess of only 0.08 rad = 4.6°Does not work as well for all inputs

AssessmentMechanical System

Sturdy apparatus key to entire project

Model of Dynamic Tilt ParametersWorks well after initial calibration of mass

Webcam Image ProcessingGood response but prone to inaccuracies

Disturbance RejectionFunctions well, but with large settling time for large disturbances

Pouring ControlOccasionally spills due to stationary cup

Overall Success!

Conclusion

Areas of ImprovementThird degree of motion Faster and more precise cup volume feedbackImplement fuzzy logic controller based on constant webcam dataImprove models and parameter identification