Analyzing and improving the reproducibility of shots taken by a soccer robot

Analyzing and improving the reproducibility of shots taken by a soccer robot

Overview Introduction Capacitor analysis Lever positioning analysis Validation Conclusion and recommendations

Introduction Shots of TURTLE not reproducible Variation capacitor voltage Leads to variation in force exerted Variation -> reproducibility Duty cycle (KK) End position Desired Variation

Introduction Lever setting control (In)accuracy Variation -> reproducibility Lever setting (LL) End position Desired Accuracy Mean Variation

Research questions What is the variation in discharge of the capacitor over the coil and what is the effect due to this factor on the reproducibility of shots? What is the variation and accuracy of the positioning of the lever, what is the effect of these factors on the reproducibility and accuracy of shots and how can this be improved?

Capacitor analysis Voltage measured by logging Voltage before shot small variation, discharge large variation Model of duty cycle (KK) versus discharge voltage (UU dddd ) 450 400 350 300 250 U c [V] 200 150 100 50 Voltage Voltage at start discharge Voltage after discharge 0 0 50 100 150 200 250 300 350 Measurement time [s]

Capacitor analysis Variation in UU dddd to find bounds of KK KK [-] UU dddd mmmmmmmm [VV] UU dddd vvvvvvvvvvvvvvvvvv [VV] 0.2 96.16 ±3.27 0.5 318.10 ±4.99 0.8 427.91 ±8.46 KK [-] KK mmmmmm [VV] KK mmmmmm [VV] 0.2 0.194 0.204 0.5 0.495 0.511 0.8 0.751 0.846

Trajectory determination Interested in influence on shots KK and LL coupled Model of relation between KK and LL versus initial angle (αα 0 ) and velocity (vv 0 ) vv 0 = cc 1 + cc 2 KK + cc 3 LL + cc 4 KKKK + cc 5 KK 2 + cc 6 LL 2 (1) αα 0 = cc 7 + cc 8 KK + cc 9 LL + cc 10 KKKK (2)

Trajectory determination Trajectory off point mass with initial velocity and angle yy = vv 0 cccccc(αα 0 )tt zz = vv 0 ssssss αα 0 tt + 1 2 ggtt2 + zz 0 yy = vv 0 cccccc(αα 0 ) zz = vv 0 ssssss(αα 0 ) + gggg

Trajectory determination End position for inconsitent KK = 0.2, 0.5 and 0.8 with a constant LL = 0 tttt 1 with steps of 0.2 Largest position error

End position variation due to KK KK [-] Largest variation of end position (yy) due to variation of KK [cm] 0.2 8 0.5 13 0.8 34 Variation in capacitor discharge voltage should be lowered, as it causes an end position error of three times the ball s radius! Larger variation at higher KK

Lever analysis Current lever positioning Not reaching reference position Spike in controller output

Lever analysis End position error of inconsistent and inaccurate LL = 0 to 1 with steps of 0.2 at constant KK = 0.2, 0.5 and 0.8 Variation up to one time radius of the ball Inaccuracy up to three times radius of the ball KK [-] Largest variation of end position (yy) due to variation of LL [cm] Largest inaccuracy of end position (yy) due to variation of LL [cm] 0.2 ±2 5 0.5 ±4 9 0.8 ±5 30

Lever analysis Improved reference position No spike Better following behavior

Capacitor analysis 1000 mmmm to study steady-state error Required controller output when lever is at a standstill (due to gravity and/or friction) Model as a feed-forward dependent on LL

Capacitor analysis 8 th degree polynomial Second degree power function

Capacitor analysis Error window to 500 counts

Capacitor analysis KK [-] Largest variation of end position (yy) due to variation of LL [cm] Largest inaccuracy of end position (yy) due to variation of LL [cm] 0.2 ±1 1 0.5 ±1 1 0.8 ±2 2 Acceptable Only smart part of ball s radius

Validation Video measurements

Validation Video measurements Setting Mean end position [m] LL = 1, KK = 0.5 (1) 8.68 ±47 LL = 0.2, K = 0.8 (2) 4.85 ±93 LL = 0.4, KK = 0.8 (3) 8.81 ±137 Variation in end position [cm] Setting End position at lower bound [m] 1 8.73 8.91 2 4.47 4.97 3 7.67 8.26 End position at higher bound [m]

Validation Bad results Plotting 2 nd degree polynomial zz versus tttttttt zz = gg 22 tt2 + vv 0 ssssss αα 0 tt + zz 0 (1) zz = gg (2) Setting Mean gg 2 [mm/ss2 ] Variation gg 2 [mm/ss2 ] 1-4.81 ±0.19 2-4.90 ±0.84 3-4.38 ±0.89 Model validated before -> still correct results

Conclusion and reccommendations Reproduciblity and accuracy improved, but reproduciblity not yet optimal Variation in discharge capacitor Feed-forward model for every TURTLE Ball handling not optimal Duty cyle and lever setting coupled Testing with control on during shot

Questions?