Space Robotics Guidance, Navigation and Control Challenges

Transcription

1 Space Robotics Guidance, Navigation and Control Challenges Jurek Sasiadek Department of Mechanical & Aerospace Engineering Carleton University Ottawa, Ontario, Canada 1

2 Presentation Outline Space manipulators - introduction - fleible joint control - fleible link control Mobile robots in Space Flying robots 2

3 Space manipulators vs Mobile robots Manipulators Mobile robots (rovers, autonomous ground Flying and floating robots (UAV and spacecraft) 3

4 Space-based Robots space robotic manipulators o can perform repetitive and lengthy tasks with reduced risk and improved performance o require less infrastructure than manned systems no life support systems application eamples o maintenance, repair, and assembly o spacecraft deployment and retrieval o etravehicular activity support 4 o shuttle inspection

5 Space-based Robots fleible links and joints o operational control challenges due to fleible effects, especially in the joints o link and joint fleibility effects introduce vibrations that can lead to instability when neglected in the control system design objectives o develop and validate advanced control systems for fleible joint space robotic 5 manipulators

6 Space-based Robots unresolved issues o simple and efficient algorithms that account for practical limitations have yet to be developed o applicability of eisting control strategies to real-time space applications (no gravity, highly limited computational load) needs to be assessed o performance validation should be done with large square trajectories (fleible effects are more noticeable greater control challenge) 6

7 Space-based Robots propose 4 control strategies for fleible joint space robots compare their respective performance while tracking a 12.6 m 12.6 m square trajectory further Improvements o etend for robots with unknown/variable joint stiffness coefficients 7

8 RMS, SSRMS and SPDM 8

9 Remote Manipulator System (RMS) 9

10 Remote Manipulator System (RMS) Canadarm1 or Shuttle arm 15.3 m long Weight 408 kg Diameter 38 cm Payload 29,500 kg 10

11 Mobile Servicing System MBS SSRMS SPDM 11

12 Space Robotic System 12

13 Space Robotic System 13

14 ISS JAXA JEM Module 14

15 Mobile Base System Technical Specs Total Length Mass (appro) Mass Handling/Transportation Capacity 5.7 m 4.5 m 2.9 m 1,450 kg 20,900kg Degrees of Freedom Peak Power (Operational) Average Power (Keep alive) Fied 825 Watts 365 Watts 15

16 Mobile Base Platform (MBS) 16

17 Space Station Remote Manipulator System (SRMS) SSRMS Arm Specifications: Width 2.2M Length 17.6M Mass (appro.) 1,800Kg Mass handling capacity 100,000Kg Degrees of freedom 7 Peak power 2000 Watts Average power 1360 Watts Stopping distance ( ma. load) 0.6m 17

18 SSRMS 18

19 Space Station Remote Manipulator System SSRMS 19

20 SSRMS folded 20

21 Special Purpose Detrous Manipulator (SPDM) Length Mass (appro) Mass Handling Capacity Degrees of Freedom 15 Peak Power Average Power Stopping Distance (under ma. load) 3.5 m 1,662 kg 600 kg 2000 Watts 600 Watts 0.15 m 21

22 SPDM - Detre 22

23 SPDM 23

24 RMS Robot 24

25 Shuttle Arm (RMS) 25

26 Space Robotics 26

27 27

28 Space Robotics 3 28

29 Satellite retrieval 29

30 Satellite robot (SPDM Detre from CSA) Tumbling Satellite Problem 30

31 Contet Space Robotic Manipulators Represent an ideal technology to perform repetitive and lengthy tasks Require less infrastructure than humans (such as life support systems) Application Eamples Maintenance, repair and assembly Spacecraft deployment and retrieve Etravehicular activity support Shuttle inspection Image courtesy of ESA ICINCO 2012 Rome, Italy, July

32 Contet Manipulators Operational control challenges due to fleible effects, especially in the joints Joint fleibility effects introduce vibrations and can lead to instability when neglected in the control system design Main Objective Develop and validate advanced control systems for fleible joint space robotic manipulators ICINCO 2012 Rome, Italy, July

33 Two-Link Space Robot Rigid Dynamics where τ = M q) q C ( q, q ) q r ( + r M M M and 2 2 ( l1 + lc2 + 2l1l c2 q2 ) + I1 I 2 2 r = m 11 1lc1 + m2 cos + 2 ( lc2 + l1l c2 q2 ) 2 r = M 12 r = m 21 2 cos + I 2 r = m 22 2lc2 + I 2 q 2 q 1 + q C r ( q, q ) = m2l1l c 2 sin q2 q ICINCO 2012 Rome, Italy, July 2012 I 33

34 Two-Link Space Robot Fleible Joint Dynamics Derived by including the kinetic energy of the rotors and considering the elastic potential energy of the linear springs at the joints M J m r ( q) q + C q m + k( q r m ( q, q ) q = q) = τ k( q m q) The link dynamics and the motor dynamics Eqs. are only coupled by the elastic torque term k( q q) ICINCO 2012, Rome, Italy, July 2012 m 34

35 Fleible Joint Control Survey Fleible Joint Control Categories Proportional Derivative (Tomei, 1991) Singular Perturbation-Based (Spong, 1989) Integral Manifold (Ghorbel and Spong, 1992) Feedback Linearization (De Luca, 1998, 2005, 2008) Optimal (Merabet and Gu, 2008) Adaptive (Slotine, 1987, 1988, 2008) Simple Adaptive Control (Sasiadek, Ulrich, Barkana, 2009, 2010, 2011, 2012) Robust (Lee, Yeon, Park and Yim, 2006, 2007) Nonlinear Backstepping (Brogliato, 1995, 1998) Fuzzy and Neural Network (Zeman, 1989, 1997) Iterative (Wang, 1995) ICINCO 2012, Rome. Italy. July

36 Fleible Joint Control Survey Unresolved Issues Simple and efficient algorithms considering practical limitations are yet to be developed Applicability of eisting control strategies to real-time space applications (no gravity, highly limited computational load) needs to be assessed Performance validation should be done with large square trajectories (fleible effects are more noticeable greater control challenge) ICINCO 2012, Rome. Italy, July

37 Advanced Control Strategies Compare 4 control strategies for fleible joint space robots Slotine and Li controller PD controller Singular Perturbation-Based controller Nonlinear Backstepping controller Compare their respective performance while tracking a 12.6 m 12.6 m square trajectory 37

38 38 Slotine and Li Control Strategy Slotine J. J. E. and Li W., On the Adaptive Control of Robot Manipulators, International Journal of Robotics Research, Vol. 6, No. 3, pp , where Advanced Control Strategies s K q q q C q q M τ d r r r r r + = ), ( ) ( ( q) q Λ q q + = c r c ( q) q Λ q q + = c c r r c c q q q q Λ q q s = = ) ( ) ( = d K positive constant gain matri

39 Advanced Control Strategies PD Control Strategy Tomei P., A Simple PD Controller for Robots with Elastic Joints, IEEE Transactions on Automatic Control, Vol. 36, No. 10, pp , τ = K p ( qmc qm) K where q mc = q c d q K = positive constant gain matri p K = positive constant gain matri d m 39

40 Advanced Control Strategies Singular Perturbation-Based Control Strategy Spong M. W., Adaptive Control of Fleible Joint Manipulators, Systems and Control Letters, Vol. 13, pp , where τ = τ s + τ f where τ s = τ r = SLI control algorithm for space robots defined earlier τ f = K v ( q q m) K = positive constant gain matri v ICINCO 2012, Rome, Italy, July

41 Advanced Control Strategies Nonlinear Backstepping Control Strategy Brogliato B., Ortega R. and Lozano R., Global Tracking Controllers for Fleible Joint Manipulators: A Comparative Study, Automatica, Vol. 31, No. 7, pp , 1995 where [ q 2( q q ) 2( q q ) ( s + s) ] + k( q q) τ = J m mc m mc m mc m 1 q k τ + q mc = r 1 q k τ + q mc = r 1 q k τ + q mc = r s = ( q q ) Λ( q q ) c k = joint stiffness matri c = q q r Note: The joint acceleration can be obtained by inverting the link dynamics equation and the jerk is obtained by time-differentiating the acceleration ICINCO 2012, Rome, Italy, July

42 Advanced Control Strategies Simulation Results Slotine and Li PD ICINCO 2012, Rome, Italy, July

43 Advanced Control Strategies Singular Perturbation-Based Nonlinear Backstepping ICINCO 2012, Rome, Italy, July

44 Novel Adaptive Control Scheme Further Improvement The Singular Perturbation-Based strategy must be etended for robots with unknown/variable joint stiffness coefficients The idea is to replace the rigid term (SLI) with an adaptive control algorithm τ = τ s + τ f As a first step, a novel model reference adaptive control (MRAC) scheme for rigid joint space robots is presented 44

45 Novel Adaptive Control Scheme Adaptive Jacobian Scheme where J(q) = Jacobian matri e T τ = J( q) K p ( t) + K ey T [ e e ] = [( ) ( y y) ] T y ref T [ e e ] = [( ) ( y y )] T y ref ref ref d e ( t) e y = l1 cos q1 + l2 cos( q1 + q2) y = l1 sin q1 + l2 sin( q1 + q2) ref c = y ref y c = s ωn 2 2ζωns + ωn 45

46 Novel Adaptive Control Scheme where the proportional adaptive gain is adapted as K with K ( t) = K pp ( t) + K pi ( t dt p ) pp ( e = 2 Γpp t) 2 0 ey 0 Γ pp K pi ( t) = e 2 Γ pi δ K 0 p pi11 ( t) e 2 y Γ pi 0 δ K p pi22 ( t) 46

47 Novel Adaptive Control Scheme Simulation Results 47

48 Space Robotic Missions to Mars Viking Lander Missions Pathfinder/Sojourner Mission (Pathfinder landed July 4-th, 1997) Spirit / Opportunity Mission Curiosity Mission (will land on August 6, 2012) 48

49 Mars robotic missions 49

50 Viking Lander Missions (Orbiter 1 and Orbiter 2) 50

51 Phoeni Lander 51

52 Spirit / Opportunity and Pathfinder / Sojourner Rovers 52

53 NASA Spirit Rover itinerary 53

54 NASA Mars Rover Opportunity Mission image taken between Dec. 21, 2011, and May 8,

55 Image taken by NASA Opportunity robot 55

56 Image taken by Opportunity robot 56

57 Images taken by Opportunity robot 57

58 Space Robots (Phoeni lander and Curiosity mobile robot) 58

59 Curiosity Mars Rover 59

60 Curiosity landing site 60

61 Mobile space robots design

62 Curiosity, Spirit and Sojourner 62

63 Guidance, Navigation and Control Autonomous robot Outdoor navigation Indoor navigation Structured environment Unstructured environment Structured environment 63

64 Mapping Mapping static local dynamic global 64

65 Local versus global Mapping Computational constraints metric topological Data association probabilistic connectivity points 65

66 Probabilistic methods Kalman filtering methods Occupancy grid mapping Particle filters SLAM Adaptive Kalman filtering Adaptive occupancy grid methods Monte Carlo method 66

67 Sensor/Data Fusion Sensor fusion is a measurement integration procedure Sensor fusion is one of the most important elements of robot GNC system 67

68 Sensor Fusion Sensor Fusion Probabilistic methods Least square methods Intelligent methods 68

69 Collision avoidance Cluttered environment Static Dynamic Proportional navigation Constant angle navigation Mapping and active collision avoidance 69

70 Trajectory Tracking Localization Positioning Coordinate frames Relative Absolute Sensors Internal Eternal Sensor fusion 70

71 Mobile robot navigation through two identifiable points Gate passage problem Navigation with vision system 71

72 ARGO 66 Conquest (Ontario Drive & Gear Limited Inc ) 6 wheel drive 4 cycle engine with 200 HP 617 cc and load capacity of 700 lbs electronic ignition 72

73 1.4 Equipped Sensors NovAtel GPS, MicroStrain inertial sensor, Built in wheel odometry, LMS-221 laser scanner (SICK) 73

74 Gate Through Navigation 74 ICINCO 2012, Rome.Italy, July 2012 Jerzy Sasiadek Location (GPS) Heading (INS)

75 Gate Recognition 75

76 Solution for Gate Through Navigation Zones of Gate Area 76 ICINCO 2012, Rome. Italy, July 2012

77 77 Visibilities of Gate Segments

78 Laser Sensor Scanning A full scan of 180 provides 361 range values 78 (indeed according to a scanning angle) ICINCO 2012, Rome, Italy, July 2012

79 Flowchart of gate ARGO Position from GPS/INS navigation module Data reading Model Start Raw Map Building Scanning Data from LMS Map Filtering Gate segments Fitting Unepected Error Check Y Mission Abort N Standard Signatures Create Current Signature (2-norm)<e N Y Localization Gate through Control ICINCO

80 Control System PGATE C T PGATE PENRTE O Y a X PENRTE 80 c

81 Parking Control / Astolfi-controller or d ρ / dt = V cosα dα / dt = V sin α / ρ ω dψ / dt = ω V = K ρρ ω = Kα + Kψ 1 2 In -y coordinate V K y = + ρ 2 2 y π ω = K (arctan( ) θ) + K ( θ)

82 Controller for Gate Through Navigation dρ / dt = V sinψ sin( α + ψ) dα / dt = V sinψ cos( α + ψ) / ρ ω dψ / dt = ω After linearization: dα / dt = K α K ψ + K ψ 1 2 dψ / dt = K α K ψ 1 2 The control law is: ω K1α K2ψ ρ = The gains are eperimentally set to K : = 0.1; K = ρ 1 0.9; K = α is estimated online by sensing. 82

83 Navigation Eperimental results 30 ARGO Trajectory Start point 15 Y (meter) End point X (meter) 83

84 Gate Through Navigation Eperimental results ARGO 12 Y (meter) GATE X (meter) 84

85 Gate Through Navigation Eperimental results 85

86 Eperimental results 2 Video 3 Video 4 86

87 Particle Filtering to estimate the robot pose: o measurements (scanning) from the SICK sensor are compared with the map scanning results robot pose estimation by particle filtering is suitable for this problem: o a set of hypotheses (particles) about the robot s pose or about the locations of the objects around the robot are maintained 87

88 Particle Filtering particle filtering algorithm is divided into four steps: o initial sampling; o prediction; o update; o re-sampling 88

89 Simulations 4.1 The robot running along the desired patrolling path is simulated. It is assumed that the robot has the pose information from GPS/Odometry sensor and the LRS supplies the scan data. Also, the a priori map of the environment is supposed to be known and is used to obtain the reference scan data. The robot starts from a point in the bottom edge and runs in anti-clockwise direction

90 Localization with Particle Filtering 50 The distribution of particles Y (meter) X (meter) Simulation of particle filtering initial distribution 50 The distribution of particles Y (meter) X (meter) Simulation of particle filtering re-sampling 90

91 ocalization with Particle Filtering 50 The distribution of particles Y (meter) X (meter) Simulation of particle filtering re-sampling 50 The distribution of particles Y (meter) X (meter) Simulation of particle filtering re-sampling 91

92 Mobile Robot Navigation Using Vision objective o achieve accurate rover navigation in crowded environments o accurate positioning, localization, and mapping o fusion of odometry, inertial navigation unit (INU), and vision data, 92

93 Robot Navigation Using Vision cameras are carried by robots o can cameras be used for accurate navigation? robot position and pose come from rotation and translation of fied camera key initial operation o identification and localization of points-of-interest 93

94 Navigation with vision system 94

95 Navigation with vision images - correspondences 95

96 UAV Navigation Using Vision one camera o motion can be derived from geometrical considerations if camera view is of points on a plane two cameras o image depth can be derived o rotation and translation can be derived using iterative closest point algorithm 96

97 Autonomous Vehicle Using SLAM Objective o to develop an efficient navigation method for autonomous robots based on SLAM and Rao- Blackwellised particle filtering with the following considerations Data association Computational compleity Non-linear motion and sensor models 97

98 Simultaneous Localization and what does the environment look like? where am I? Mapping Basics where am I supposed to go? how can I get there? 98

99 A Comparison between EKF-SLAM and Fast-SLAM 100 EKF-SLAM with Gaussian EKF-SLAM with Nom-Gaussian Implication Assumption Fast-SLAM with Non-Gaussian Assumption y direction [m] Estimated position error y direction [m] Position Estimated position error along y ais [m] Time (s) direction [m] EKF-SLAM with Nom-Gaussian Implication Time (s) [s] Position error along y ais [m] 99 Position estimation error direction [m] Fast-SLAM with Non-Gaussian Implication Time (s) Time [s]

100 FAST-SLAM Vs. EKF-SLAM in a Dense Map for Multiple Objects (750 Landmarks) EKF-SLAM FAST-SLAM EKF Performance for multiple objects True Path Dead Reckoning Estimated Path (EKF) RBPF performance for multiple objects True Path Dead Reckoning Estimated Path (RBPF) y direction [m] y direction y direction [m] y direction direction [m] 10 direction [m] direction direction 100

101 Fast-SLAM vs. EKF-SLAM in a Dense Map for Multiple Objects (750 Landmarks) 50 Mapping Detectable multipleobjects landmarks by RBPF EKF True map y direction y direction direction direction 101

102 FAST-SLAM Vs. EKF-SLAM in a Dense Map for Multiple Objects (750 Landmarks) Mapping Detectable multipleobjects landmarks by RBPF EKF True map y direction y direction direction direction

103 Conclusions Space robotics includes many control challenges; Present applications of robots in Space are not fully autonomous; Full autonomous operations should be our objective. 103

104 Follow your Curiosity! 104

105 Acknowledgement Images courtesy of Canadian Space Agency and NASA or from sites: www. csa.ca

106 Thank you! Questions? 106