Soft Robotics Control Strategies for Robots Interacting with Humans

Alin Albu-Schffer

together with: Christian Ott, Sami Haddadin, Andreas Stemmer, Thomas Wimbck

DLR German Aerospace Center Institute of Robotics and Mechatronics

Service Robotics

Our Vision
Affordable, remotely controlled operations in space with mobile (freeflying) robonauts for Servicing and Exploration

Mechatronic Joint Design

(redundant)

Control components
Light-weight robot with elastic joints Joint torque sensor Movement accuracy Safe human-robot-environment interaction

active vibration damping

compliance control collision reaction

robust task execution self-collision avoidance

8

Vibration Damping
Light-weight Measurement higher joint compliance vibrations

torque sensor

vibration damping

Joint Level Control

, ()

d d

D
controller

, ()

q
ext
passive environment

passive controlled actuators - eq. (2),(4)

rigid robot dynamics eq.(1)

4th order state feedback & & controller: , , , same structure used for

toque control position control impedance control

10

11

Control components
Light-weight robot with elastic joints Joint torque sensor Movement accuracy Safe human-robot-environment interaction

active vibration damping

compliance control collision reaction

robust task execution self-collision avoidance

12

Cartesian Stiffness Control

Kk M

f x

xd

_ _ _

xd

Dk

13

Cartesian Impedance Controller

Two step concept for noncollocated systems:
4 Shaping the potential energy - collocated feedback
Asymptotic stabilization around xd ( ) Implementation of the desired compliance relationship ( Feedback of )

4 Shaping of the kinetic energy - noncollocated feedback

Damping of vibrations => increased performance Feedback of (torque controller)

=> Full state feedback

14

Main Idea for Energy Shaping xd fext x

At equillibrium: 1 to 1 correspondence

Between A controller based on is collocated instead of

and

passivity

satisfies static requirements related to q: - desired equilibrium point - desired stiffness

15

Cartesian Impedance Control Unified approach for torque, position and impedance control on Cartesian and joint level
Passive environment

xd

x(q) J(q)

kinematics dynamics

q() h-1() m B , B ext k d a

g(q)
impedance law

torque control

q
rigid robot dynamics

motor dynamics Passive structure

16

Impedance Control

17

18

Impedance Behavior for Two-handed Manipulation

19

Robust Assembly Strategy

Object localization
(Image Processing)

Geometrical Analysis

Automatical generation of optimal assembly trajectory Impedance based execution of planed trajectory

Maximal robustness w.r.t. position and form errors

20

Collision Detection

robot

motor dynamics
& q

rigid body dynamics

ext

& q

disturbance observer

ext ext

21

Collision Detection

F m
motor dynamics
& q

robot

rigid body dynamics

ext

, &
q
& q, q
disturbance observer

ext ext

22

Impact Experiments

strategy 1: stopping the trajectory strategy 2: gravity compensated torque mode

investigated criteria for head, neck, chest

23

Safety aspects
How to define a safety standards for robots?

New standard for safety in Robotics: ISO 10218 says:

4 v_max < 0.25m/s OR 4 F_max < 150N OR 4 P_max = 80W

24

How dangerous is the robot really?

25

HIC - Head Injury Criterion

For all tested criteria the LWR was in the lower quarter of the green area

26

Conclusions
Under assumption of joint elasticity, the torque becomes a state full state feedback Passivity based control is particularly suited for unknown environments Unified approach for position, torque and impedance control Impedance control needs adequate robust planning strategies Collision detection based on torque sensors is more accurate The DLR robot is safe (from automotive point of view) even at 2m/s We need new injury indices customized for robotics for protecting humans AND robots
27

Placing of Pedicle Screws

28

Experimental Setup with LBR II

Robot Linear guides with marker arrays

Navigation system

Registration of vertebra

Linear guides with marker arrays

29

Conditions for Energy Shaping

In any equilibrium position q=q() if k not too small For very small k:

q1

q2 -second equilibium

fixed
k d For a general potential energy U

One turn
diffeomorphism
30

Uniqueness of the solution Invertibility

q()

Extendable to a broad class of noncollocated E-L systems

Control approaches
Light-weight robot with elastic joints Joint torque sensor Movement accuracy Safe human-robot-environment interaction

active vibration damping

compliance control collision reaction

robust task execution self-collision avoidance

31