Motion generation for humanoid robots aiming at industrial - - PowerPoint PPT Presentation

▶
motion generation for humanoid robots aiming at
SMART_READER_LITE
LIVE PREVIEW

Motion generation for humanoid robots aiming at industrial - - PowerPoint PPT Presentation

Nov 18, 2022 120 likes •439 views

LAAS-CNRS Laboratoire conventionn avec lUniversit Fdrale de T oulouse Midi-Pyrnes Motion generation for humanoid robots aiming at industrial applications June 21-22 th 2017, Robotex, TechDays 2017, Clermont-Ferrand O. Stasse,

slide-1
SLIDE 1

LAAS-CNRS / Laboratoire d’Analyse et d’Architecture des Systèmes Laboratoire conventionné avec l’Université Fédérale de T

  • ulouse Midi-Pyrénées

Motion generation for humanoid robots aiming at industrial applications

  • O. Stasse, Gepetto

Team, LAAS-CNRS June 21-22th 2017, Robotex, TechDays 2017, Clermont-Ferrand

slide-2
SLIDE 2

Table of Contents

1 Motivations 2 A humanoid platform for industrial applications 3 Control architecture 4 Conclusions & Perspectives

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 2/24 Motion generation for humanoid robots aiming at industrial applications

slide-3
SLIDE 3

Acknowledgements

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 3/24 Motion generation for humanoid robots aiming at industrial applications

slide-4
SLIDE 4

Testing our algorithms in real industrial use case Safety Power

Humanoid Robot for the factory

  • f the future

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 4/24 Motion generation for humanoid robots aiming at industrial applications

slide-5
SLIDE 5

A replacement to HRP-2

  • Pros
  • Highly reliable
  • Very robust (not able to

jump ...)

  • Understanding of the

software architecture

  • Cons
  • Closed access to the low

level part

  • DC - Brush motors ( unable

to achieve fast speed- high torque)

  • Low software support 10.04

LTS · · ·

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 5/24 Motion generation for humanoid robots aiming at industrial applications

slide-6
SLIDE 6

Characteristics for a new humanoid

  • Should hold a 10Kg tool at

strech arm

  • Reaction to push recovery

(need high speed - high torque)

  • Frequency of control at 1KHz

for OSID

  • Low level access needed
  • Skin ?

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 6/24 Motion generation for humanoid robots aiming at industrial applications

slide-7
SLIDE 7

The robots around

  • ATLAS - Hydraulic actuation -

2 M$ - Not available

  • WALKMAN - Electric

actuation - Support is an issue

  • Valkyrie - SEA - 2.7 M$ -

Need a contract agreement with NASA (Edinburgh)

  • TORO - Not available -

Maybe one day through Kuka

  • Proxi SRI - 97 % transmission

efficiency

  • REEM-C - Position control AC

motor - HD - 300 k euros

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 7/24 Motion generation for humanoid robots aiming at industrial applications

slide-8
SLIDE 8

The robots around

  • ATLAS - Hydraulic actuation -

2 M$ - Not available

  • WALKMAN - Electric

actuation - Support is an issue

  • Valkyrie - SEA - 2.7 M$ -

Need a contract agreement with NASA (Edinburgh)

  • TORO - Not available -

Maybe one day through Kuka

  • Proxi SRI - 97 % transmission

efficiency

  • REEM-C - Position control AC

motor - HD - 300 k euros

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 7/24 Motion generation for humanoid robots aiming at industrial applications

slide-9
SLIDE 9

The robots around

  • ATLAS - Hydraulic actuation -

2 M$ - Not available

  • WALKMAN - Electric

actuation - Support is an issue

  • Valkyrie - SEA - 2.7 M$ -

Need a contract agreement with NASA (Edinburgh)

  • TORO - Not available -

Maybe one day through Kuka

  • Proxi SRI - 97 % transmission

efficiency

  • REEM-C - Position control AC

motor - HD - 300 k euros

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 7/24 Motion generation for humanoid robots aiming at industrial applications

slide-10
SLIDE 10

The robots around

  • ATLAS - Hydraulic actuation -

2 M$ - Not available

  • WALKMAN - Electric

actuation - Support is an issue

  • Valkyrie - SEA - 2.7 M$ -

Need a contract agreement with NASA (Edinburgh)

  • TORO - Not available -

Maybe one day through Kuka

  • Proxi SRI - 97 % transmission

efficiency

  • REEM-C - Position control AC

motor - HD - 300 k euros

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 7/24 Motion generation for humanoid robots aiming at industrial applications

slide-11
SLIDE 11

The robots around

  • ATLAS - Hydraulic actuation -

2 M$ - Not available

  • WALKMAN - Electric

actuation - Support is an issue

  • Valkyrie - SEA - 2.7 M$ -

Need a contract agreement with NASA (Edinburgh)

  • TORO - Not available -

Maybe one day through Kuka

  • Proxi SRI - 97 % transmission

efficiency

  • REEM-C - Position control AC

motor - HD - 300 k euros

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 7/24 Motion generation for humanoid robots aiming at industrial applications

slide-12
SLIDE 12

The robots around

  • ATLAS - Hydraulic actuation -

2 M$ - Not available

  • WALKMAN - Electric

actuation - Support is an issue

  • Valkyrie - SEA - 2.7 M$ -

Need a contract agreement with NASA (Edinburgh)

  • TORO - Not available -

Maybe one day through Kuka

  • Proxi SRI - 97 % transmission

efficiency

  • REEM-C - Position control AC

motor - HD - 300 k euros

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 7/24 Motion generation for humanoid robots aiming at industrial applications

slide-13
SLIDE 13

Pyrene: First prototype of the TALOS serie

[Submitted to IHRC 2017] https://www.youtube.com/watch?v=SxdNvP2jKcc

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 8/24 Motion generation for humanoid robots aiming at industrial applications

slide-14
SLIDE 14

Motion generation: the problem

min f (u(t), v(t)) g(u(t), v(t)) < 0 h(u(t), v(t)) = 0

      

f(t): The cost function u(t): The control vector g(t): The inequality constraints h(t): The equality constraints v(t): The environment model

t CoM ˆ q Balance (under-actuated part) GIK A general problem on the time window

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 9/24 Motion generation for humanoid robots aiming at industrial applications

slide-15
SLIDE 15

SDKs

  • Gepetto is developping Software Development Kits
  • Stack Of Tasks
  • Humanoid Path Planner
  • Try to identify software patterns from the problem

formulation

  • Write our own solvers when needed (often)
  • Be as much generic as possible
  • Fragment the code
  • Integration through a build farm with binairies (robotpkg)

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 10/24 Motion generation for humanoid robots aiming at industrial applications

slide-16
SLIDE 16

Motion generation

t CoM u Balance (under-actuated part) GIK

∆T ∆T ∆T ∆T ∆T ∆T ∆T ∆T ∆T ∆T Free-flyer u1 u n

2

un

A general problem on the time window

Size of the problem 2 × 200 × 30 = 9600 variables Non linear constraints Discrete nature due to contacts

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 11/24 Motion generation for humanoid robots aiming at industrial applications

slide-17
SLIDE 17

MuJoCo [Koenemann, IROS 2015]

https://www.youtube.com/watch?v=WbsQBPzQakc

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 12/24 Motion generation for humanoid robots aiming at industrial applications

slide-18
SLIDE 18

Motion generation

min f (u(t), v(t)) g(u(t), v(t)) < 0 h(u(t), v(t)) = 0

      

t CoM ˆ q Balance (under-actuated part) GIK A general problem on the time window

Planning and control solve the same problem Planning is looking for a global feasible solution Control is looking for on online sensor grounded local solution Planning is too long when simulating the control Control can fails Local minima leading to an incomplete behavior Mismatch between the control and the hardware Accessibility set [Majumdar, ICRA Best Paper Award 2013]

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 13/24 Motion generation for humanoid robots aiming at industrial applications

slide-19
SLIDE 19

HPP [IJRR Submitted]

http://stevetonneau.fr/files/publications/ijrr16/video.mp4

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 14/24 Motion generation for humanoid robots aiming at industrial applications

slide-20
SLIDE 20

Take away messages

  • The embodiment (mechanical body, limits and controllers)

defines the motion capabilities of the robot.

  • We need to connect the accessibility set of our controllers to

the planner.

  • We need an efficient computation of the mechanical

quantities

  • We need to break down the problem complexity with small

but representative problems

  • We need to push higher the semantic level of our motion

controllers

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 15/24 Motion generation for humanoid robots aiming at industrial applications

slide-21
SLIDE 21

Motion generation: the constraints

min f (u(t), v(t)) g(u(t), v(t)) < 0 h(u(t), v(t)) = 0

      

t CoM ˆ q Balance (under-actuated part) GIK A general problem on the time window

M1(q)¨ q + N1(q, ˙ q)˙ q + G1(q) = T1(q)u + C⊤

1 (q)λ Actuated dynamics of the robot

M2(q)¨ q + N2(q, ˙ q)˙ q + G2(q) = C⊤

2 (q)λ

Underactuated dynamics of the robot

f (λ) ∈ F

General balance criteria

umin < u < umax

Torques limits

ˆ qmin < ˆ q < ˆ qmax

Joints limits

d(Bi(q), Bj(q)) > ǫ, ∀p(i, j) ∈ P

(self-)collisions

¨ ei = ˙ Ji(q)˙ q + Ji(q)¨ q

Tasks

          

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 16/24 Motion generation for humanoid robots aiming at industrial applications

slide-22
SLIDE 22

Motion generation: the constraints

Pattern generator Focus on the underactuated part Model predictive control

t CoM ˆ q Balance (under-actuated part) GIK A general problem on the time window

M1(q)¨ q + N1(q, ˙ q)˙ q + G1(q) = T1(q)u + C⊤

1 (q)λ Actuated dynamics of the robot

M2(q)¨ q + N2(q, ˙ q)˙ q + G2(q) = C⊤

2 (q)λ

Underactuated dynamics of the robot

f (λ) ∈ F

General balance criteria

umin < u < umax

Torques limits

ˆ qmin < ˆ q < ˆ qmax

Joints limits

d(Bi(q), Bj(q)) > ǫ, ∀p(i, j) ∈ P

(self-)collisions

¨ ei = ˙ Ji(q)˙ q + Ji(q)¨ q

Tasks

          

Simplifying the walking problem to con- trol only the CoM reference

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 16/24 Motion generation for humanoid robots aiming at industrial applications

slide-23
SLIDE 23

Motion generation: the constraints

Inverse dynamics Focus on the inertia matrix Forces Complete constraints

t CoM ˆ q Balance (under-actuated part) GIK A general problem on the time window

M1(q)¨ q + N1(q, ˙ q)˙ q + G1(q) = T1(q)u + C⊤

1 (q)λ Actuated dynamics of the robot

M2(q)¨ q + N2(q, ˙ q)˙ q + G2(q) = C⊤

2 (q)λ

Underactuated dynamics of the robot

f (λ) ∈ F

General balance criteria

umin < u < umax

Torques limits

ˆ qmin < ˆ q < ˆ qmax

Joints limits

d(Bi(q), Bj(q)) > ǫ, ∀p(i, j) ∈ P

(self-)collisions

¨ ei = ˙ Ji(q)˙ q + Ji(q)¨ q

Tasks

          

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 16/24 Motion generation for humanoid robots aiming at industrial applications

slide-24
SLIDE 24

Under actuation

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 17/24 Motion generation for humanoid robots aiming at industrial applications

slide-25
SLIDE 25

Generalized locomotion Climbing stairs

[Kudruss, Humanoids 2015] [Carpentier, ICRA 2016] Previous work [Luo, ICRA 2014] [Vaillant, Humanoids 2014] [Noda, ICRA 2014] [Hirukawa, ICRA 2007]

m(¨ c(u) − g) = Nc

i=1 fi, λi = [f⊤ i µ⊤ i ]⊤

mc(u)×(¨ c(u) − g) + w⊤(u)Iw⊤(u) = Nc

i=1 pi×fi + µi

    

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 18/24 Motion generation for humanoid robots aiming at industrial applications

slide-26
SLIDE 26

Whole body motion

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 19/24 Motion generation for humanoid robots aiming at industrial applications

slide-27
SLIDE 27

Multi-contacts with Pyrene

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 20/24 Motion generation for humanoid robots aiming at industrial applications

slide-28
SLIDE 28

Motion control: Instantaneous control

Inverse dynamics Focus on the inertia matrix Forces Complete constraints

t CoM ˆ q Balance (under-actuated part) GIK A general problem on the time window

M1(q)¨ q + N1(q, ˙ q)˙ q + G1(q) = T1(q)u + C⊤

1 (q)λ Actuated dynamics of the robot

M2(q)¨ q + N2(q, ˙ q)˙ q + G2(q) = C⊤

2 (q)λ

Underactuated dynamics of the robot

f (λ) ∈ F

General balance criteria

umin < u < umax

Torques limits

ˆ qmin < ˆ q < ˆ qmax

Joints limits

d(Bi(q), Bj(q)) > ǫ, ∀p(i, j) ∈ P

(self-)collisions

¨ ei = ˙ Ji(q)˙ q + Ji(q)¨ q

Tasks

          

Ignore the actuator dynamics

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 21/24 Motion generation for humanoid robots aiming at industrial applications

slide-29
SLIDE 29

Actuator dynamics

Classical actuator SEA - Romeo - [IROS 2017] Mc Kibben Muscle - [IROS 2016]

φ(u) = τj

Actuators dynamics

M1(q)¨ q + N1(q, ˙ q)˙ q + G1(q) = T1(q)τj + C⊤

1 (q)f Actuated dynamics of the robot

M2(q)¨ q + N2(q, ˙ q)˙ q + G2(q) = C⊤

2 (q)f

Underactuated dynamics of the robot

f (f) ∈ F

General balance criteria

umin < u < umax

Torques limits

ˆ qmin < ˆ q < ˆ qmax

Joints limits

d(Bi(q), Bj(q)) > ǫ, ∀p(i, j) ∈ P

(self-)collisions

¨ ei = ˙ Ji(q)˙ q + Ji(q)¨ q

Tasks

              

Actuator dynamics: necessity to control

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 22/24 Motion generation for humanoid robots aiming at industrial applications

slide-30
SLIDE 30

Torque control with HRP-2

  • A. Del Prete, T. Flayols

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 23/24 Motion generation for humanoid robots aiming at industrial applications

slide-31
SLIDE 31

Conclusions & Perspectives

We are reaching the limits of our actuators They need to be taken into account in our control architecture New mechanical design are needed Reaching the functionnal level of human may involve robot looking less than human

LAAS-CNRS /Laboratoire d’Analyse et d’Architecture des Systèmes LAAS-CNRS - Genova 2017 - April, 19th - O. Stasse 24/24 Motion generation for humanoid robots aiming at industrial applications