Improving autonomous orchard vehicle trajectory tracking performance - PowerPoint PPT Presentation

Improving autonomous orchard vehicle trajectory tracking performance via slippage compensation Dr. Gokhan BAYAR Mechanical Engineering Department of Bulent Ecevit University Zonguldak, Turkey This study was conducted under the Supervision of Dr. Marcel Bergerman in the Field Robotics Center of Robotics Institute of Carnegie Mellon University, Pittsburgh, PA, USA.

Objective of the Research Development of a slippage estimation procedure and performing a desired trajectory tracking control. 1

a single set of controller parameters or a unique equation of motion to guarantee a desired performance and accuracy Due to changing the characteristics Of wheel-ground interaction 2

the simple assumptions which are generally used in the mobile robot / autonomous vehicle applications: • ideal transmission • ideal rolling • no slippage • no lost of traction control • no external wheel forces • no surface change behavior • no disturbance, etc. 3

Forward Desired task Vehicle Velocity <Mobile Robot> Model [f(x,y,t)] Unmanned Steering Ground Vehicle Controller [f(x,y)] Angle x,y, θ , V, δ Wheel-Ground Surface Interaction Information 4

Trajectory Tracking Control of an Autonomous Vehicle X Error (t)= |X Desired (t) - X Actual (t)| Y Error (t)= |Y Desired (t) - Y Actual (t)| θ Error (t)= | θ Desired (t) - θ Actual (t)| Vehicle desired (t) Vehicle actual (t) f(X,Y,t) desired Y f(X,Y,t) actual X 5

Desired Trajectory Generator � Dynamic approaches � Kinematic/Car ‐ like robot approach � Point mass model � Dubins curves 6

7 Car ‐ like robot model

8

Desired trajectory tracking controller X desired X e Σ x, y + V V c - Vehicle Φ Controller X PS Φ c Y desired Y e Φ Σ V + - Y PS θ desired θ e Σ + - θ PS 9

10 Lyapunov Functions

Working Environment of an Orchard Robot Vehicle w 1 trees w 2 11

Reference Trajectory 120 100 80 60 40 20 0 -20 -40 -20 0 20 40 60 80 100 120 12

Turning Geometry 13

14

Experimental Orchard 15

1. Experiments to test the behaviour of the proposed model � Slippage information is not taken into consideration. � RTK-GPS is used for position feedback. 16

4 km autonomous drive achieved in the orchard 17

Desired and actual steering angles for 4 km autonomous drive 18

Video 19

2. Experiments to test the behaviour of the proposed model. � Slippage information is not taken into consideration. � Row Detection System (via Laser Scanning RangeFinder) is used. 20

21

22

23

24

25

Experimental results obtained in the first row of the orchard Experimental results obtained in the first row. Width = 4.44 m, Length = 52.95 m. (a) Steering angles, (b) Lateral errors 26

Video ‐ First Row Width = 4.44 m, Length = 52.95 m 0.5 m/s Forward Velocity Forward Camera Front Camera 27

3. Experiments to test the behaviour of the proposed model. � Slippage information is taken into consideration. � RTK-GPS is used for position feedback. 28

Odometer RTK-GPS Steering System 29

30

Car Like Robot Model Without Slippage Car Like Robot Model With Slippage It is assumed that 31

Slippage Experiments on Snow 32

Reference Trajectory Tracking Control on Snow 6 Desired 4 Real 2 Vehicle Y-Direction [m] 0 Control -2 Without Slip -4 Estimation -6 -8 -10 0 5 10 15 20 X-Direction [m] 6 Desired 4 Real Vehicle 2 Control Y-Direction [m] 0 With Slip -2 Estimation -4 -6 -8 -10 0 5 10 15 20 33 X-Direction [m]

30 w/o Estimation 20 w/ Estimation 10 Steering Angle [deg] 0 -10 -20 -30 -40 1.4 -50 0 20 40 60 80 w/o Estimation Time [s] w/ Estimation 1.2 Forward Speed [m/s] 1 0.8 0.6 0.4 0.2 0 0 20 40 60 80 34 Time [s]

4. Orchard Experiments. � Slippage information is taken into consideration. � Row Detection System (via Laser Scanning RangeFinder) is used. 35

36

37

E1 � results obtained by using RTK GPS feedback without using slippage estimation. E2 � results obtained by using the slippage estimation procedure that uses RTK GPS feedback. E3 � results obtained by using feedback information coming from dead reckoning algorithm. No slippage estimation procedure is adapted into the system model. E4 � results obtained by using the slippage estimation process that uses the dead reckoning feedback information. 38

Video ‐ Turning control without slippage estimation 39

Video ‐ Turning control with slippage estimation 40

Special thanks to the co-authors of the paper: Gokhan Bayar*, Marcel Bergerman 1 , E. ilhan Konukseven 2 , A. Bugra Koku 2 , “Improving the trajectory tracking performance of autonomous orchard vehicles using wheel slip compensation”, Biosystems Engineering , vol. 146, pp. 149- 164, 2016. 1 Field Robotics Center, Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA 2 Mechanical Engineering Department, Middle East Technical University, Ankara, Turkey

Thanks for your attention