iCub Whole-Body Control Through Force Regulation on Rigid - - PowerPoint PPT Presentation

▶
icub whole body control through force regulation on rigid
SMART_READER_LITE
LIVE PREVIEW

iCub Whole-Body Control Through Force Regulation on Rigid - - PowerPoint PPT Presentation

Mar 31, 2024 181 likes •546 views

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts Francesco Nori IIT Robotics, Brain and Cognitive Sciences iCub Whole-Body Control iCub Whole-Body Control Through Force Regulation on Through Force Regulation

slide-1
SLIDE 1

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-2
SLIDE 2

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-3
SLIDE 3

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

WHY

3

slide-4
SLIDE 4

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

WHY

4

slide-5
SLIDE 5

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-6
SLIDE 6

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-7
SLIDE 7

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-8
SLIDE 8

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

WHAT

8

slide-9
SLIDE 9

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-10
SLIDE 10

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

WHAT

10

slide-11
SLIDE 11

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

HOW

11

⇥ J(q) J(q) ⇤  ˙ vq ˙ vq

  • + ˙

J(q, vq) + ˙ J(q, vq) = 0 M(q)˙ vq + C(q, vq)vq + g(q) =  τq

  • + J>(q)f

M(q)˙ vq + C(q, vq)vq + g(q) =  τq

  • +
  • + J>(q)f

Self Other Contact

slide-12
SLIDE 12

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

HOW

12

slide-13
SLIDE 13

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Technology

HOW

13

  • G. Metta, L. Natale, G.Sandini, F

. Nori

Science

  • S. Traversaro, F

. Romano, D. Pucci, L.Fiorio, A. Del Prete, J. Eljaik, F . Nori

Software

RBCS: S. Traversaro, J. Eljaik, ICUB: A. Cardellino, D. Domenichelli, G. Metta, L. Natale ADVR: A. Rocchi, M. Ferrati, E. Mingo Hoffman, A. Settimi, N. Tsagarakis, A. Bicchi

slide-14
SLIDE 14

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Science

  • S. Traversaro, F

. Romano, D. Pucci, L.Fiorio, A. Del Prete, J. Eljaik, F . Nori

Software

RBCS: S. Traversaro, J. Eljaik, ICUB: A. Cardellino, D. Domenichelli, G. Metta, L. Natale ADVR: A. Rocchi, M. Ferrati, E. Mingo Hoffman, A. Settimi, N. Tsagarakis, A. Bicchi

Technology

HOW

14

  • G. Metta, L. Natale, G.Sandini, F

. Nori

slide-15
SLIDE 15

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Technology

15

slide-16
SLIDE 16

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

HOW

16

  • U. Pattacini, F. Nori, L. Natale, G. Metta, G. Sandini (2010).

International Conference Intelligent Robots and Systems (IROS).

Dynamics Kinematics

IPOPT library for inverse kinematics with nonlinear constraints (reach, gaze, foot placement, c.o.m. control). Joint torques estimation from embedded proximal force/ torque sensing. NE whole-body internal and external force estimation fusing force and tactile sensing.

Compliance

  • J. Eljaik, Z. Li, M. Randazzo, A. Parmiggiani, G. Metta, N.

Tsagarakis, F. Nori (2014). Robotics: Science and Systems (RSS).

  • M. Fumagalli, S. Ivaldi, M. Randazzo, L. Natale, G. Metta, G.

Sandini, and F. Nori (2012). Autonomous Robots (AURO).

slide-17
SLIDE 17

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Joint Torques and External Wrenches Estimation

  • Goal: measure robot interaction with the world
  • Effect of motors : joint torques
  • Interaction with environment: external wrenches
  • We have a specific set of sensors on the iCub
  • skin, six axis Force Torque sensors, IMU
  • We exploited the classical RNEA, applied to the robot

subtrees induced by the F/T sensors

Fumagalli, Nori et Al. (2012). Force feedback exploiting tactile and proximal force/torque sensing. Autonomous Robots. Measured wrench Estimated joint torque Estimated external wrench

Whole-body Tactile sensors Six-axes force/torque sensors

slide-18
SLIDE 18

Francesco Nori

Robotics, Brain and Cognitive Sciences

Humanoid whole-body motion control with distributed contacts

  • we want to improve actuators disturbance rejection
  • movements in presence of unpredictability
  • active feedback might not be a viable strategy
  • we implemented part of the control loop mechanically
  • passive mechanical feedback
  • we proposed a novel actuator design
  • agonist-antagonist design based on non-linear springs

passive noise rejecting VSA

  • L. Fiorio, F. Romano, A. Parmiggiani, G. Sandini, F. Nori (2014). Robotics, Science and Systems (RSS).
  • the actuator joint has a unique equilibrium position
  • antagonistic activation of motors: co-contraction
  • co-contraction increases passive noise rejection
slide-19
SLIDE 19

Francesco Nori

Robotics, Brain and Cognitive Sciences

Humanoid whole-body motion control with distributed contacts

  • we constructed a first prototype
  • non-linear spring based on two specialized cam designs
  • complex components have been 3D printed
  • we derived two control action
  • open-loop position and stiffness control
  • open-loop stochastic optimal control

passive noise rejecting VSA

  • F. Romano, L. Fiorio, G. Sandini, F. Nori (2014). IEEE Multi-conference on Systems and Control (MSC).
slide-20
SLIDE 20

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-21
SLIDE 21

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Technology

HOW

21

  • G. Metta, L. Natale, G.Sandini, F

. Nori

Science

  • S. Traversaro, F

. Romano, D. Pucci, L.Fiorio, A. Del Prete, J. Eljaik, F . Nori

Software

RBCS: S. Traversaro, J. Eljaik, ICUB: A. Cardellino, D. Domenichelli, G. Metta, L. Natale ADVR: A. Rocchi, M. Ferrati, E. Mingo Hoffman, A. Settimi, N. Tsagarakis, A. Bicchi

slide-22
SLIDE 22

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Open source (control) software

22

slide-23
SLIDE 23

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Control Software

23

  • S. Traversaro, A. Del Prete, S. Ivaldi, F. Nori (2015). IEEE

International Conference in Robotics and Automation (ICRA).

  • J. Eljaik, A. del Prete, S. Traversaro, M. Randazzo, F. Nori (2014).

Workshop at the International Conference in Robotics and Automation (ICRA). Mingo H.E., Traversaro S., Alessio R., Mirko F., Settimi A., Romano F., Natale L., Bicchi A., Nori F., Tsagarakis G. (2014). Modelling and Simulation for Autonomous Systems.

iDynTree wholeBodyInterface Gazebo

A YARP-based open source simulator for the iCub, COMAN and Walkman robot. A generic library for dynamics computation, used to model articulated rigid bodies. A software abstraction layer for whole-body motion control and dynamic modelling.

slide-24
SLIDE 24

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Software

RBCS: S. Traversaro, J. Eljaik, ICUB: A. Cardellino, D. Domenichelli, G. Metta, L. Natale ADVR: A. Rocchi, M. Ferrati, E. Mingo Hoffman, A. Settimi, N. Tsagarakis, A. Bicchi

Technology

HOW

24

  • G. Metta, L. Natale, G.Sandini, F

. Nori

Science

  • S. Traversaro, F

. Romano, D. Pucci, L.Fiorio, A. Del Prete, J. Eljaik, F . Nori

slide-25
SLIDE 25

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Basic science

25

slide-26
SLIDE 26

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Control Software

26

D.Pucci, F. Romano, F. Nori (2015). Submitted to Transaction on Robotics (TRO).

  • B. Berret, E. Chiovetto, F. Nori, T. Pozzo (2011). PLoS Comput

Biology.

Adaptive control Hierarchic optimal control Muscle modelling

M o d e l s o f m u s c l e c o - activation and its use in compensating disturbances. Adaptive control of under- a c t u a t e d m e c h a n i c a l systems. Novel formalism to plan

  • ptimal movements with

hierarchic costs.

  • F. Romano, A. Del Prete, N. Mansard, F. Nori (2015). Robotics and

Automation (ICRA).

A prototype for compliant actuator with muscle like properties.

  • L. Fiorio, F. Romano, A. Parmiggiani, G. Sandini, F. Nori (2014).

Robotics: science and systems (RSS).

  • F. Nori, B. Berret, L. Fiorio, A. Parmiggiani, G. Sandini (2013). EP

13785613.4 .

slide-27
SLIDE 27

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

The control architecture 27

  • Interaction control calls for torque control
  • Two nested loops
  • Inner loop stabilises desired joint torques
  • Outer loop generates desired joint torques

“iCub Whole-Body Control through Force Regulation on Rigid Noncoplanar Contacts”, Frontiers in Robotics (2014)

  • Inner loop
  • identification of friction and motor parameters
  • Outer loop
  • Impedance/inverse dynamics/adaptive control
slide-28
SLIDE 28

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

The inner loop

slide-29
SLIDE 29

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

  • Goal: identify the gap between CAD and reality
  • Extracting parameters from CAD is hard
  • CAD models may be incorrect
  • We exploited the sensors in iCub to perform inertial

parameters identification

  • We used the classical base link technique, applied to the

robot subtrees induced by the F/T sensors

  • S. Traversaro, A. Del Prete, S. Ivaldi, F. Nori. (2015) IEEE International Conference on Robotics and Automation (ICRA).

Theory: Results:

Measured wrench Identified parameters: Masses Center of Masses, Inertia Matrices :

Inertial Parameters Identification

slide-30
SLIDE 30

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

The outer loop: balancing 30

“Force Control of the iCub Humanoid for One Foot balancing and Safe Interaction”, ICRA and Humanoids 2015

  • External wrenches controlled to stabilise a desired robot’s momentum
  • Redundancy of external wrenches used for joint torque minimisation
  • Actuation redundancy used to stabilise a desired robot’s “internal” configuration
  • Dynamics
  • Constraints

M(q) ˙ ν + h(q, ν) − J>(q)w = Sτ

JF (q)ν = 0

slide-31
SLIDE 31

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

slide-32
SLIDE 32

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

The outer loop: balancing on seesaw

  • External wrenches on the robot controlled to stabilise a desired robot’s momentum
  • Redundancy of external wrenches used for controlling seesaw and joint torque minimisation
  • Actuation redundancy used to stabilise a desired robot’s “internal” configuration
  • Robot Dynamics
  • Seesaw dynamics
  • Constraints

M(q) ˙ ν + h(q, ν) − J>(q)w = Sτ JF (q)ν = VF VP = 0

˙ Hs = mg − Aw + wP

slide-33
SLIDE 33

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

The outer loop: adaptive control 33

  • Collocated dynamics depends nonlinearly from inertial parameters (Slotine inapplicable)
  • No need of acceleration measurement
  • Asymptotic stability of reference trajectory in the sense of Lyapunov
  • System Dynamics
  • Control objective
  • Adaptive control: unknown inertial parameters

M(q) ˙ ν + h(q, ν) − J>(q)w = Sτ

qj → r(t)

“Collocated Adaptive Control of Underactuated Mechanical Systems”, Transaction On Robotics, 2015

slide-34
SLIDE 34

/34

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts

Hierarchical Differential Dynamic Programming34

  • Introduce strict priorities in Optimal Control [1]
  • Avoid task weight tuning (as in Prioritized Control)
  • Consider the whole trajectory (as in Optimal Control)
  • [2] Improves efficiency by taking inspiration from

Differential Dynamic Programming (Mayne 1966)

[1] A. Del Prete, F. Romano, L. Natale, G. Metta, G. Sandini, F. Nori (2014). International Conference on Robotics and Automation (ICRA).
 [2] F. Romano, A. Del Prete, N. Mansard, F. Nori (2015). International Conference on Robotics and Automation (ICRA).

Each iteration of the algorithm is composed of:

  • Problem approximation. The original nonlinear problem is approximated along a nominal trajectory

into a linear dynamical system and a quadratic cost.

  • Backward pass. An optimal control solution is computed for the local linear approximation.
  • Forward pass. The computed control solution is applied to the nonlinear system. A line search

procedure is also applied. It ensures that the iteration leads to a proper descent step for the original nonlinear problem.

slide-35
SLIDE 35

Francesco Nori

IIT Robotics, Brain and Cognitive Sciences

iCub Whole-Body Control Through Force Regulation on Rigid Non-Coplanar Contacts