Antonio Bicchi & Phriends · Colloquium on R&A -Napoli18 Dec., 2006 © Università di Pisa...

© Università di PisaColloquium on R&A - Napoli 18 Dec., 2006

Physical Human-Robot Interaction:Dependability, Safety, and Performance

Antonio Bicchi& Phriends

Centro “E. Piaggio”University of Pisa


Popular notions of robotics have long foreseen humans and robots existing side-by-side sharing work

Until very recently, the reality has been quite different

Industrial robots have been far too dangerous to share space with humans, while physical interaction has been far too unintuitive.

Introduction


Currently industrial robots work in far more structured environments than is required for human workers


pHRI: Robots in touch with HumansShortly, all this is going to change radically:the next generation of robots will interact with people directly.

• cooperative manipulation tasks• domestic applications (domotics)• entertainment• assistance• rehabilitation

SARCOSNASA-JPL

The Lokomat by Hocoma


UTAH

• teleoperation • human augmentation• haptic exoskeletons• mixed environments

pHRI: Robots in TOUCH with Humans

Berkeley

HAL – U. Tsukuba


Accuracy less demanding Safety is a must

Keep using rigid robots;Increase sensors drastically;Use active control;

An approach: Design for Accuracy, Control for Safety

Modern approach: Design for Safety, Control for AccuracyMechanical (passive) compliance;Compensation by control;

Towards Intrinsically Safe Mechanisms

• Machines interacting with humans have different requirements than current in industry


Research Issues for pHRI

Design

Sensors

Software

Control

Biomimetics

Planning

LighrtweightCompliance

LighrtweightCompliance

SafetyPerformance

SafetyPerformance

On-Line FusionDependability

On-Line FusionDependability

InterfacesHuman metrics

InterfacesHuman metrics

Open ArchitectureDependability

Open ArchitectureDependability

Real TimeConsistency

Real TimeConsistency


• The “holy grail” of pHRI design is intrinsic safety:

• to design a robot that will be safe for humans no matter what failure, malfunctioning, or even misuse might happen.

• Naturally, perfect safety against all odds is not feasible for machines which have to deliver performance in terms of weight lifting, swift motion, etc.:

• the trade-off between safety and performance is the name of the game in pHRI.

Safety in pHRI


Traditional approaches:modify controllers for rigid robot manipulators

(stiffness, impedance control)add sensors (force, contact, proximity,…)

Well known intrinsic limitations to alter by control the behavior of the arm if the mechanical bandwidth is not matched to the task

In other words, making a rigid, heavy robot to behave gently and safely is an almost hopeless task, if realistic conditions are taken into account.

co-design of mechanics and control

Safety in pHRI


What does ‘safety’ mean? • The engineering field that has developed broadest experience with human safety in collisions is car industry;

• “Severity indices” are used as measures of risk in studies in passive damage in car accidents [Versace, 1971];

• Zinn et al. [2001] first used the Head Injury Coefficient (HIC) in robotics

• (other robotics-specific measures?) Tmax

Vope

t


Conventional Design

• Given rotor + link inertia, covering, and acceptable HIC max. velocity

InertiaM

ax sa

fe sp

eed

Iso-HIC curve

v

Operator Inertia

CompliantCovering

RotorInertiaRotorInertia

LinkInertiaLink

Inertia

• First: minimize rotor + link inertia, use compliant covering if possible


The generation of DLR light-weight robots is among the most advanced implementations of the “traditional”approach

The DLR-III Light-Weight Arm lifts a payload equal to its own weight (13,5 kg).

Justin: 2 L-W Arms with Torso(DLR + UNINA)

Safety-oriented design


Collision detection and reaction (DLR+UOR)

• Collision detection

• Collision reaction strategies

• Stop

• Switch to Grav. Comp.

• Swich to Grav. Comp and additional link inertia shaping

• Admittance control Flee

• Trajectory scaling Go back and forth along commanded trajectory

• Adaptive trajectory deformation

• …


Theoretically, HIC goes to zeroas Kf increases.

Conventional Design with Active Force Control- Ideally

Unsafe

Safe

Operator Inertia

TotalInertiaTotalInertia K InterfaceK Position }

Kf

Pos. Ref. v

Force feedback

+v

Force Controller GainH

IC


Crash Tests (DLR + KUKA + ADAC)


Crash Tests (DLR + KUKA + ADAC) Instrumentation


Conventional Design with Active Force Control- Really

1) limited torque/link inertia ratio

2) limited mechanical bandwidth

3) limited sampling bandwidthH

IC .Safe

Unsafe

Ratio of Link Inertia vs. Max. Torque

Active force control unsafe in real conditions

Sampling timeForce feedback Operator Inertia

TotalInertiaTotalInertia K InterfaceK Position }

Kf

Ref. Pos. v+

v


Conventional Design with Active Force Control- Really

HIC

Experimental Data


How to get beyond these limitations?

uCompliantCoveringRotor

InertiaRotorInertia

LinkInertiaLink

Inertia


InertiaRotorInertia

LinkInertiaLink

Inertia

Basic idea: decouple rotor inertia from link

via mechanical elasticity

Heavy Light


Soft Robots

Spring Flamingo (MIT Leg Lab)

Knee Orthosis (Yobotics)

VUB - SoftArm R&MM

COG


How to get beyond these limitations? - 2


InertiaRotorInertia

LinkInertiaLink

Inertia


InertiaRotorInertia

LinkInertiaLink

Inertia

Basic idea: decouple rotor inertia from link

via mechanical elasticity

Heavy Light

Controllable Mechanical Compliance of the Joint


Soft Robotics

Constant Passive Compliance

Variable Passive Compliance

•Compliance is fixed•Absorb impact effects•Adapt by changing the elastic element•Only one motor

•Compliance can be changed•Adapt to situations, e.g.:

• position a heavy load• move a glass of liquid

•Extra motor to alter compliance•Increased complexity


(Slowly) Variable Stiffness Arms

AMASC (Hurst 2004)

Mechanical Impedance Adjuster (T. Morita, 1995)


Variable Stiffness Actuation

Quickly Variable Compliance

• Compliance can be changed with time constants comparable to motion• Adapt stiffness during tasks, e.g.:

•Robot arm: compliant in quick motion vs. stiff accurate end positioning•Legged locomotion: adapt to gaits and swing/stance phase

• (Larger) extra motor to alter compliance

Slowly Variable Compliance

• Compliance can be changed once per task • Adapt to nominally planned tasks• Extra motor to alter compliance


(Quickly) Variable Stiffness Arms

Compliant StiffVUB R&MM MACCEPA (R. Van Ham 2006)

UNIPI VSA-1 (2002)

Video


New VSA Design

Video 1: Principle

Video 2: Soft mode

Video 3: Stiff mode


What does ‘Performance’ mean?

•A concern – not unique – is promptness of response:

How long does it take to bring the arm to a prescribed position?• endless actuator authority perfectly safe in zero time

• different controllers produce different performance and risks for the same mechanism

• decouple mechanical performance from controloptimal control: the Safe Brachistocrone


Safe Brachistochrone for VSA

The Safe Brachistocrone optimal control problem

The control of transmission compliance can recoverperformance

LinkInertia

RotorInertia

vrot vlinkuK Kcov

xrot xlink

uact

Safety & Control boundsVST Bounds


Simulation results for VSA (rotor & link)

LinkInertia

RotorInertia

vrot vlinkutransm Kcov

xrot xlink

uact

The philosophy: Fast & Soft, Stiff & SlowH

ICTime


Actuator for VSA: Stiffness Variation

Video


Actuator for VSA: Stiffness Variation

Video

HIC results in the ramp accelerating phase for VSA


Extending to more DOF’s

Impact at the elbow joint

Impact at the end effector

HIC values are calculated off-line by simulationat varying configuration, stiffness and velocity’s direction (constant module)

Analytical expression for the HIC is non-trivialin this case


Extending to more DOF’s

Impact at the elbow joint

Impact at the end effector

HIC values are calculated off-line by simulationat varying configuration, stiffness and velocity’s direction (constant module)

Analytical expression for the HIC is non-trivialin this case

impact at the end of a link is not necessarily the worst case


Impacts of a 2 DOF VSA arm

Elbow

End-effector

q2= 90o

q1 = 0


Impacts of a 2 DOF VSA arm

5000


Allowable range for directionsof the stiffness

ellipsoid minor axis

Stiffness Ellipsoid and HIC LocusStiffness ellipsoid: plotted by evaluating the eigenvalues and

eigenvectors of the matrix

The direction of the stiffness ellipsoid changes at varyingjoint stiffness

11

2001 −−−−⎟⎟⎠

⎞⎜⎜⎝

⎛==Σ J

KK

JKJJ TT


Vee (safer)

Vee (dangerous)

Stiffness Ellipsoid and HIC Locus

HIC locus describes the injury level in case of impacts at the end-effector at varying the joints positions, stiffnessand direction of the e-e velocity. (compare with “belted inertia ellipsoids”).


Stiffness Ellipsoid and HIC Locus

There is an intimate correlation between maximum HIC, maximum stiffness, and effective inertia at the e-e


Trajectory/Stiffness Planning for a 2 DOF robot armFor a point-to-point minimum time/safe motion, the safe brachistochronealgorithm tends to align the minimum stiffness eigen-direction with (scaled) velocity

vv

vv


Conclusions

• From stiff, heavy robots towards lightweight, soft robots

• Safety indices and tests• The variable stiffness approach • Towards n-dof’s• Much remains to be done:

– New hardware design– New planning problems– New control strategies


Physical Human-Robot Interaction:Dependability, Safety, and Performance

Antonio BicchiThanks to:

Giovanni ToniettiRiccardo SchiaviSoumen Sen

Centro “E. Piaggio”University of Pisa

Antonio Bicchi & Phriends · Colloquium on R&A -Napoli18 Dec., 2006 © Università di Pisa...

Documents

Transcript of Antonio Bicchi & Phriends · Colloquium on R&A -Napoli18 Dec., 2006 © Università di Pisa...