Gregory Bock, Brittany Dhall, Ryan Hendrickson, & Jared Lamkin Project Advisors: Dr. Jing Wang & Dr. In Soo Ahn Department of Electrical and Computer

Transcription

1 Gregory Bock, Brittany Dhall, Ryan Hendrickson, & Jared Lamkin Project Advisors: Dr. Jing Wang & Dr. In Soo Ahn Department of Electrical and Computer Engineering April 26 th, 2016

2 Outline 2 I. Introduction II. E-puck Brittany III. Kilobot - Jared IV. QBot 2 Ryan & Greg V. Summary & Conclusions

3 I. Introduction 3

4 Objectives 4 Design and Experimental Validation of Cooperative Control Algorithms Sensing/communication between robots Implementation of local flocking control algorithms Implementation of local formation control algorithms

5 Project Background 5 Cooperative systems found in nature Flock of birds School of fish Swarm of insects

6 Possible Applications 6 Cooperative systems found in engineering Smart Grid Sensor Network Traffic Network orate/ wildpoldsried.php

7 Heterogeneous Groups 7

8 Heading Alignment 8

9 Point Consensus 9

10 Following

11 Design Constraints 11 Must overcome limited communication among networked robots Must overcome limited sensing capability of robots Must overcome system uncertainties

12 Test Platform Kilobot 12 Diameter of 3.3 cm Two differential vibration motors IR transmitter and receiver (7 cm range) Ambient light sensor

13 Test Platform E-puck 13 Diameter of 7 cm IR transmitter and receiver ring (25 cm range) On-board CMOS camera Bluetooth 2.0 dspic 30F6014A on-board computer

14 Test Platform QBot 2 14 Open-architecture autonomous ground robot Xbox 360 Kinect Kobuki robot base Gumstix DouVero Zephyr on-board computer

15 II. E-puck Brittany 15

16 Work Accomplished 16 Software & hardware implementation Object detection/following Odometry Vicsek Model Fix battery issues

17 Infrared proximity sensors 17 8 infrared proximity sensors Composed of two parts -IR emitter & photo-sensor Can detect objects within 4 centimeters IR7 IR0 IR6 IR1 IR5 IR2 IR4 IR3

18 Object Detection and Following 18 Proximity sensors -> detected distance Compare with true specified distance Velocity = gain*(specified distance detected distance)

19 Object Detection 19

20 Object Following 20

21 Odometry 21 Using odometry the E-puck can compute their position and orientation x = S cos θ(k) + θ 2 y = S sin θ(k) + θ 2 θ k + 1 = θ(k) + θ 3 S average change in steps of both left and right motors θ change in the angle of the agents heading

22 Vicsek Model 22 θ i k + 1 = θ n i k + j=1 n + 1 θ i k Next heading of agent θ i k - current heading of agent n j=1 θ j (k) θ j (k) - sum of all neighboring agents at time k n - number of neighboring agents

23 Vicsek Model 23

24 E-puck Battery Problem Solution 24 Original Design Solution #1 Solution #2 Bad connection between positive and negative terminals from battery to E-puck Added an addition on top of E-puck, for better connection to terminals Resoldered positive terminal

25 Testing communication between E- 25 puck and Kilobot Tested E-puck communications with infrared receiver connected to oscilloscope initially, followed by testing with Kilobot Verified E-pucks sent message with correct protocol Verification of communication between E-puck and Kilobot would be accomplished by observing change in LED from red to green

26 Infrared Receiver Circuit 26 38kHz Infrared Receiver Module Infrared Receiver Circuit Used a 5V supply & oscilloscope to view the signals 330 Ω 5 V DC SIGNAL

27 III. Kilobot - Jared 27

28 Kilobot 28 Atmega MHz) 32kB flash, 1kB EEPROM, 2kB SRAM 2 vibration motors IR LED and receiver Ambient light sensor

29 How Kilobots Communicate 29 Use infrared light Measures light intensity to calculate distance Messages are sent every 200 milliseconds

30 Color Synchronization Video 30

31 Kilobot Movement: Orbiting 31 Zone of Attraction Zone of Orientation Zone of Repulsion Kilobot Kilobot

32 Multiple Agent Orbiting 32

33 Simple Localization: Gradient 33 Can determine how many agents are displaced from a specified agent Individuals receive gradient values from local agents until a buffer is full Smallest value in buffer is incremented by 1, which becomes agent s gradient value

34 Gradient 34

35 Advanced Behaviors 35 By combining gradient, orbiting, and/or light detection more advanced behaviors can be achieved such as: Fixed-point consensus: Kilobots converge to a fixedpoint Edge following: Kilobots orbit multiple stationary agents Follow-the-leader

36 Fixed-Point Consensus 36

37 Edge-Following 37

38 Follow-the-Leader 38

39 III. QBot 2 Ryan & Greg 39

40 QBot 2 - Ryan 40

41 Non-linear Model 41 Non-linear Model x = vcos θ y = vsin θ θ = ω

42 Linear Model 42 Linear Model p x = u x p y = u y p x = x + l cosθ p y = y + l sinθ

43 Simulink Model 43

44 Simulink Model 44

45 Localization 45 Color Detection Depth Calculation Communication

46 Localization Color Detection col

47 Localization Depth Calculation α d col

48 Localization Depth Calculation 48 α = 320 column π/180 α is obtained angle column is the array column number P x = d P y = d tan(α) d is depth

49 Localization Communication 49

50 Point Consensus Control Algorithm 50 u ix t = k i u iy t = k i n j=1 n j=1 Communication Topology s ij t = s ij t p jx t p ix t s ij t p jy t p iy t

51 Point Consensus 51 Communication Topology s ij t =

52 Point Consensus 52

53 Y Position (m) Point Consensus 53 Point Consensus 1 QBot 49 QBot 50 QBot X Position (m)

54 Heading Alignment 54

55 QBot 2 - Greg 55

56 Object Avoidance 56 Used Fuzzy Logic Inputs taken from Xbox 360 Kinect Outputs are left and right motor velocities 0 Left Center Right

57 Object Avoidance 57 Membership Function Plots 1 Not Clear Clear mm Input Variable Right

58 Object Avoidance 58 Membership Function Plots 1 Close Middle Far mm Input Variable Center

59 Object Avoidance 59 Membership Function Plots 1 Stop Slow Medium Fast m/s Output Variable Vr

60 Object Avoidance 60 Membership Function Plots 1 -Slow Stop Slow Medium Fast m/s Output Variable Vl

61 Fuzzy Rule Set 61 Input Output Left Center Right V R V L - Far - Medium Medium - Middle - Slow Slow Not Clear Close Not Clear Slow -Slow Clear Close Not Clear Slow Stop Not Clear Close Clear Stop Slow Clear Close Clear Slow Stop

62 Object Avoidance 62 V L = k x d x + x d + V L V R = k y d y + y d + V R

63 Object Avoidance 63

64 Y Position (m) Object Avoidance 64 Object Avoidance 1 QBot X Position (m)

65 Formation Control 65

66 Formation Control 66 u ix t = k i n j=1 s ij t p jx t C jx p ix t + C ix n u iy t = k i j=1 s ij t p jy t C jy p iy t + C iy

67 Formation Control 67

68 Y Position (m) Formation Control 68 Formation Control QBot 49 QBot 50 QBot X Position (m)

69 IV. Summary & Conclusions 69

70 Problems Encountered 70 E-puck CMOS camera For communication between different platforms, additional circuity was needed Kilobot Kilobot motors need frequent calibration Small size makes it difficult for QBot 2 to detect Lack of sensory information

71 Summary & Conclusions 71 Designed cooperative control algorithms for heterogeneous groups of robots Implemented algorithms on different robot platforms

72 Future Work 72 Cross-platform communication Implement E-puck camera Further development of formation algorithms Complete E-puck to Kilobot communication Add IR messaging system to QBot 2 Improve QBot 2 algorithm to avoid objects consistently

73 Acknowledgements 73 Our group would like to thank Dr. Wang & Dr. Ahn for their support throughout the project. Our group would also like to thank Mr. Mattus and Mr. Schmidt for their technical support.

74 Gregory Bock, Brittany Dhall, Ryan Hendrickson, & Jared Lamkin Project Advisors: Dr. Jing Wang & Dr. In Soo Ahn Department of Electrical and Computer Engineering April 26 th, 2016

75 Division of Labor Overview 75 Kilobots Jared Individual Behavior QBot 2s Ryan/Greg E-pucks Brittany/Jared Kilobot - Kilobot Jared Individual Communication QBot - QBot Ryan/Greg E-puck - E-puck Brittany/Jared Kilobot - E-puck Jared/Brittany Integrated Communication Kilobot - QBot Jared/Ryan/Greg E-puck - QBot Brittany/Ryan/Greg Algorithm Design Linearization Based Model Jared/Brittany/Ryan/Greg Integrated Behavior Formation Control Behavior Jared/Brittany/Ryan/Greg Flocking Behavior Jared/Brittany/Ryan/Greg Testing Software Implementation Jared/Brittany/Ryan/Greg Hardware Implementation Jared/Brittany/Ryan/Greg

76 Algorithm Test Platforms 76 Kilobot E-Puck QBot

77 Unbricking the E-pucks 77 Uses the MPLAB ICD 3 In-circuit Debugger MPLAB IDE v8.30 Erases the Flash memory by powering the E-puck through the ICD 3

78 Changing the Original Timer 78 E-puck s clock speed is 8 times faster than the Kilobot Increased the timer of the E-puck by a factor of 8 to slow down the rate at which the message was sent to Kilobot Change was made to allow Kilobots to sync with E- puck messaging

79 Object Avoidance 79 Inputs taken from Xbox 360 Kinect 0 Left Center Right

80 Oscilloscope Screen Captures from 80 the Infrared Receiver Circuit Original timer used in initial Kilobot testing Increased original timer by a factor of 8 Decreased original timer by a factor of 8

81 Advanced Localization: Distributed 81 Trilateration Gradient is only a 1D localization Minimum of 3 fixed agents as reference points Non-localized agents assume position (0,0) Determine actual distance to non-localized agent Calculate assumed distance

82 Advanced Localization: Distributed 82 Trilateration Direction Vectors are generated from reference to unknown Generate assumed coordinates from Vectors and measured Distances New position is determined using assumed position and previous position

83 Integrated Communication Set-up 83

84 Project Platform Costs 84 Platform Quantity Total Price QBot 2 3 $9, Kilobot Kit 20 $4, Epucks 3 $5,093.00

85 Programming Software Costs 85 Software Quantity Total Price Kilobot Controller IDE 1 $0.00 E-puck Programming Software 1 $0.00 MATLAB Courseware 1 $0.00

86 E-puck Object Following Code 86

87 QBot Point Convergence Code 87

88 QBot Obtained Angle Equation 88 α = 320 column π/180

89 HIL Write Block 89

90 Find Object Parameters 90 Specify RGB values Value threshold Number of objects

91 Overall Simulink Model 91

92 Motor Control 92

93 Localization Equations 93 C i = x 0 x i 2 + y 0 y i 2 V i = < x 0 x i, y 0 y i > C i C i n i = x i, y i D i V i x 0, y 0 = x 0, y 0 (x 0 n ix, y 0 n iy ) 4

94 Color Consensus 94 Kilobots are initialized with a random number Each number corresponds to a color Kilobots then begin transmitting value Kilobots receive messages and keep track of how many neighbors are what color Kilobots then change their color to most prevalent color

95 Color and Object Detection 95 The E-puck CMOS camera is capable of 640X480 resolution, in color or grayscale However, the image is too large to process, so instead we use a 1X120 image Color uses RGB565, where each pixel has 5 bits for red, 6 bits for green, and 5 bits for blue

96 Color and Object Detection 96 First step to object detection is edge detection The image array is searched for two edges, from both left and right starting positions Individual pixels are compared to the average of the previous ten pixels If the difference is greater than three, that location is set as an edge Based on the number of edges found (0,1,2,3,4), The E-puck calculates where the center of the object is, and how wide it is.

97 Color and Object Detection 97 After Edge detection is complete, the E-puck moves on to color comparison The E-puck computes the average RGB value of the object The average is compared to the specified value within a certain tolerance If the comparison is acceptable, The E-puck begins maneuvering to it.

98 Odometry 98 θ = R L 2 S = R+ L 2 x = S cos θ + θ y = S sin θ + θ x k + 1 = x k + x y k + 1 = y k + y θ k + 1 = θ + θ 3 2 2