1

Torsten Kroeger 1/28 Stanford University

Robot Motion Control During Abrupt

Switchings Between Manipulation Primitives

Torsten Kroeger

Artificial Intelligence Laboratory

Department of Computer Science

Stanford University, USA

Bernd Finkemeyer

KUKA Laboratories GmbH

Augsburg, Germany

Torsten Kroeger 2/28 Stanford University

Acknowledgement

Institut für Robotik und Prozessinformatik, Prof. Wahl

Technische Universität Braunschweig, Germany

This research has mainly been conducted at

2

Torsten Kroeger 3/28 Stanford University

Sensor integration and sensor-based control at multiple levels

is key for the future advancement of robotic systems.

Introduction and Motivation

www.omron.com � www.jr3.com � www.theimagingsource.com � www.sick.com � ww.robotic.dlr.de � www.kuka.com � www.honda.com � www.activerobots.com � www.kinemedic.de

Torsten Kroeger 4/28 Stanford University

• In general, we distinguish between:

1. Trajectory-following motions

2. Sensor-guided motions (sensors in the feedback loops)

- Force/torque control

- Visual servo control

- Distance control

- 9

• In real-world robot applications, systems need to switch between

controllers at unforeseen instants.

• The theoretical tool for such switchings is hybrid switched-

system control.

• How does the real world look like?

Introduction and Motivation

3

Torsten Kroeger 5/28 Stanford UniversityICRA 2004

Torsten Kroeger 6/28 Stanford UniversityICAR 2003

4

Torsten Kroeger 7/28 Stanford UniversityAdvanced Robotics 2005

Torsten Kroeger 8/28 Stanford University

5

Torsten Kroeger 9/28 Stanford UniversityAdvanced Robotics 2005

Torsten Kroeger 10/28 Stanford University

Switching from One MP to Another

• Controllers cannot take over in any situation, e.g.,

- no valid sensor signal,

- force control without contact,

- sudden changes of the environment,

- sensor malfunctions, or

- switching to a controller yields heavy jerk.

• Many compliant motion tasks require instantaneous switchings

from one controller to another in order to achieve stability.

• Requirement: Autonomous switchings at the lowest possible

control level!

• Motion specification by Manipulation Primitives (MPs).

• This framework bases on

[Mason’81, Bruyninckx’96, Finkemeyer’02, Milighetti’05, Zieliñski‘10]

and specifies sensor-guided and sensor-guarded robot motions.

6

Torsten Kroeger 11/28 Stanford University

The Adaptive Selection Matrix

Po

sitio

n

Ve

locity

Dis

tan

ce

Fo

rce

/To

rqu

e

Fu

rth

er

ϕy

ϕz

ϕx

z

y

x

dof 01

2level

device

ϕy

dofP

ositio

n

Ve

locity

Dis

tan

ce

Fo

rce

/To

rqu

e

Fu

rth

er

ϕz

ϕx

z

y

x

device

Adaptive selection matrixClassical selection matrix

An example: Level 0 Level 1 Level 2

‘Force in z-Direction‘ Force Vision Velocity

Torsten Kroeger 12/28 Stanford University

Task Space Control Scheme

7

Torsten Kroeger 13/28 Stanford University

• Hybrid switched-control (simplified for one DOF)

Control Scheme

Torsten Kroeger 14/28 Stanford University

On-Line Trajectory Generation

www.reflexxes.com

Target-position-based on-line trajectory generation.

8

Torsten Kroeger 15/28 Stanford UniversityTarget-position-based on-line trajectory generation.

On-Line Trajectory Generation

www.reflexxes.com

Torsten Kroeger 16/28 Stanford University

On-Line Trajectory Generation

www.reflexxes.com

Target-velocity-based on-line trajectory generation.

9

Torsten Kroeger 17/28 Stanford UniversityTarget-velocity-based on-line trajectory generation.

On-Line Trajectory Generation

www.reflexxes.com

Torsten Kroeger 18/28 Stanford University

Instantaneous Controller Switching

10

Torsten Kroeger 19/28 Stanford UniversityIEEE TRO 2010

Torsten Kroeger 20/28 Stanford University

Instantaneous State Space Switching

11

Torsten Kroeger 21/28 Stanford University

Torsten Kroeger 22/28 Stanford University

Unforeseen Switching of Reference Frames

12

Torsten Kroeger 23/28 Stanford University

Reference Frame A Reference Frame B

Torsten Kroeger 24/28 Stanford University

Instantaneous Reactions to Sensor Events

13

Torsten Kroeger 25/28 Stanford University

• Stabilizing

switched-systems

Velocity-acceleration plane in state space:

ICRA 2010

Torsten Kroeger 26/28 Stanford University

Summary

• Manipulation Primitives as

- a general interface to hybrid switched-control systems for

robot motion control, and

- a “language” to specify sensor-guided and sensor-guarded

robot motions and tasks (� adaptive selection matrix).

• Reflexxes Motion Libraries provide on-line trajectory generation

algorithms to

- guarantee continuous jerk-limited motions during abrupt

switchings between Manipulation Primitives, and to

- provide a safe backup controller in a hybrid switched-system.

14

Torsten Kroeger 27/28 Stanford University

Acknowledgements

Literature

• Please google for on-line trajectory generation.

- T. Kröger. On-Line Trajectory Generation in Robotic Systems.

Springer STAR Vol. 58, 2010.

- T. Kröger, F. M. Wahl. On-Line Trajectory Generation: Basic

Concepts for Instantaneous Reactions to Unforeseen Events.

IEEE T-RO, 2010.

- T. Kröger. On-Line Trajectory Generation: Straight-Line

Trajectories. IEEE T-RO, 2011.

Torsten Kroeger 28/28 Stanford University

T. Kroeger, B. Finkemeyer:

Robot Motion Control During Abrupt Switchings Between

Manipulation Primitives.

Acknowledgements

Thank You for Your Attention!

15

Torsten Kroeger 29/28 Stanford University

The End!