Design and Control of Compliant Humanoids. Alin Albu-Schäffer. DLR German Aerospace Center Institute of Robotics and Mechatronics

Transcription

1 Design and Control of Compliant Humanoids Alin Albu-Schäffer DLR German Aerospace Center Institute of Robotics and Mechatronics

2 Torque Controlled Light-weight Robots Torque sensing in each joint Mature technology for experimental platforms

3 First Applications of the Technology in Automotive Industry Sensor based assembly with the LWR KUKA Demonstrator Robot commercialized by KUKA since 2008 Fast, open research software interface (FRI) Available since 2010

4 Gearbox Assembly at Daimler Special Gripper from earlier solutions Production started /7 Application with the LWR More than gearbox units in Mercedes cars Production without fences. Humans interact with the robot

5 Modularity and light-weight allows the construction of complex kinematics using the arm joints DLR crawler DLR walker [Ott & al. Humanoids 2010, 2011]

6 Mechatronic Joint Design (redundant)

7 The complete Finger System

8 The DLR-HIT-hands on the way to commercialization Hand I with four fingers,12 actuators tooth belt drives 1kg finger tip force torque control Hand II with five fingers,15 actuators

9 DEXHAND Europe s first Robonaut hand Tendon driven, 12 active dof Less than 3.3 kg Finger length 93 mm (Thumb 100 mm), DEXHAND length 340 mm 25 N Fingertip force (Thumb 40 N) streched out 25 N test

10 New light-weight robot with f/t sensing in the joints and at extremities - Higher power and speed SARAH

11 Cartesian Impedance Control Unified approach for torque, position and impedance control on Cartesian and joint level x d Impedance control d torque control K T f. B motor dynamics Robot dynamics a 1 F (1 K ) T F B B (1 K T ) 1 B Passivity Robustness in contact with the environment [Albu-Schäffer & al, IJRR 2007]

12

13 DLR Hand II Impedance Control Joint impedance Control Cartesian Impedance Control Object Impedance Control [Wimböck al. IJRR 2010]

14 Impedance Control for Two Handed Manipulation Gravity compensation d V ( ) g ( ) D( ) Stiffness term 1ms control cycle for the whole system Damping term

15 Human-Robot-Interaction Compliant Control of the entire Robot Impedance Control w Admittance Control M D f v Rollin Justin 53 active dof 150 kg

16 Current Research Plattform based on Variable Compliance Actuation (VIA) Non linear spring Load Motor 1 q1 q Motor 2 q2 Output Link Motion The antagonistic concept Stiffness adjustment

17 Anthropomorphic Hand-Arm-System Size, force and dynamics of a human arm/hand Variable stiffness 52 motors,111 position sensors No torque sensing: torque observed from positions

18 A Hand-Arm System for Space Robot Assistrance Extension of the passivity based control approaches to the VIA robots: Variable, nonlinear stiffness Strongly coupled joints

19 The new integrated hand-arm-system (with variable impedance actuation VIA) Grebenstein & al. Humanoids 2010]

20 VIA Variable Impedance Actuators 1 Antagonistic Actuator (fingers) 2 equally sized motors motors pull tendons Changing Stiffness Moving Joint Output I M1 External Load Motor 1 θ 1 Motor 2 I M2 θ 2 q Control τ L Torque Estimator

21 VIA Variable Impedance Actuators 2 Bidirectional Antagonistic Actuator (underarm rotation and wrist) 2 equally sized motors both motors push and pull (bidirectional) Moving Joint Output Changing Stiffness Motor 1 q External Load Nonlinear Springs Motor 2 [Petit & al. ICRA 2010]

22 VIA Variable Impedance Actuators 3 Adjustable Stiffness Actuator (upper arm) one big motor1 moves the joint one small motor2 changes joint stiffness without motor2 we have a serial elastic joint Motor 1 Moving Joint Output Harmonic Drive Gear External Load q Nonlinear Springs Changing Stiffness Motor 2 [Wolf & al. ICRA 2008]

23 Joint Data Sheet: DLR VS-Joint VIACTORS Actuator Type Maximum Joint Torque (repeatable, evaluated by measurement) Min./Max. Stiffness (no external load) Max. Storable Energy Max. Equilibrium Velocity Nominal Power (not max./peak!) Min. Stiffness Adjusting Time (from 3% to 97% stiffness) Torque Hysteresis at Max. Torque Weight (w/wo Motors) Size (w/wo Motors) Max. Deflection Range (min./max Stiff.) Variable Stiffness ± 180 Nm 0 / 315 Nm/rad 16.8 J 217 /s = 320 W 0.2 s 7.3% 1.4 / ~ 2.0 kg Ø97x106 / ~ Ø97x166 mm ± 14 / ± 14 φ Joint Deflection cφ c : Radius of Cam Disk

24 Joint Data Sheet: DLR QA-Joint VIACTORS Actuator Type Maximum Joint Torque (repeatable, evaluated by measurement) Min./Max. Stiffness (no external load) Max. Storable Energy Max. Equilibrium Velocity Nominal Power (not max./peak!) Min. Stiffness Adjusting Time (from 3% to 97% stiffness) Quasi Antagonistic ±40 Nm 20 / 550 Nm/rad 2.7 J 217 /s = 320 W 0.15 s Stiffness [Nm/rad] Torque Hysteresis at Max. Torque Weight (w/wo Motors) Size (w/wo Motors) Max. Deflection Range (min./max Stiff.) +/-12.5% 1.4 / ~ 2.0 kg Ø90x100 / ~ Ø90x160 mm ± 15 / ± 3 Torque [Nm]

25 Validation of Arm Robustness

26 Control of VIA Joints The joints have very low intrinsic damping useful for cyclic movements involving energy storage (running or throwing) damping of the arm for fast, fine positioning tasks has to be realized by control. Ensuring the achievement of the desired link position with motor position based control. Providing the desired stiffness property.

27 General Model For all considered actuator types so far, following model structure holds External disturbance torque Main properties: - under-actuation: less control inputs ( 1, 2 ) than dimension of configuration space - positive definiteness of V (x) We propose this generic model for controller design of VIA joints Flexible joint model is a particular case [Albu-Schäffer at ICRA 2010]

28 Decoupling in Modal Coordinates back to link coordinates K K K K P D T S QK Q K Q K Q K PQ DQ TQ SQ Q T Q Q Q T T T symmetric, nondiagonal p.d state feedback controller in link coordinates. u K ~ 1 1 P KD KT K KS K [Petit at ICRA 2010]

29 Experimental Validation Point to Point trajectory Vibration Damping OFF Vibration ON

30 Experimental Validation

31 Cartesian Impedance Control Implementation of a simple Cartesian impedance V ( q) τ m g( q) D( q) q q Potential: V ( q) VS ( H ( q), H d, d ) H d K d Damping design: Double diagonalization of the inertia matrix and the Hessian of the potential function. 6 F ext H (q) Extension for variable stiffness joints Combine active and passive impedance [Petit at IROS11]

32 Passive Joint Elasticities & Cartesian Stiffness generate by passive joint stiffness diagonal Cartesian transformation bounded [Albu-Schäffer et al. 03] severe limitation of achievable Cartesian stiffness error: 25-55%

33 Combining Active & Passive Impedances active impedance controller passive joint stiffness best of both worlds: overcome limitations of passive stiffness get VSA features

34 1. Step: Passive Compliance Optimization achieve Cartesian compliance as good as possible by passive compliance least-squares problem solution by active-set algorithm efficient (pseudo inverse) fast (366Hz for 4 joints) passive compliance residual

35 2. Step: Active Compliance Optimization remove residual by active compliance optimization formulation: possibly negative definite stability issues.. goal matrix.. positive definite solution by matrix nearness problem efficient (eigenvalue problem) [Higham 98]

36 Results motor pos link pos passive spring dominant: motion provided by link active spring dominant: motion provided by motor

37 Performance Validation Rigid joint VIA joint Optimal control for maximizing end velocity. - Analytical solutions for 1dof, linear case - Extension to nonlinear case with dynamic constraints

38 Constant vs. Variable Stiffness Increase in velocity for the QA joint [Haddadin at IFAC 2011]

39 Performance Validation for multi dof 2 dof system ball throwing Evaluation of human-inspired throwing motion generation

40 Performance Demonstration with the Hand Chalon & al. IROS 2011

41 Performance Validation: Kicking Experiments

42 Experimental Results Speed Kicking range Impact joint torque Stiff Joint 3.06 m/s 1.6 m 85 Nm VS-Joint 6.35 m/s 4.05 m 10 Nm

43 WP2- Robotics of Biological Neuro-mechanical Control Imperial College London [Ganesh & al., TRO 2011]

44 The DLR Hand-Arm System