T T I I I I I I I I T T University of Lbeck Institute of - - PowerPoint PPT Presentation

Organic Fault-tolerant Robot Control Architecture

11th Colloquium Organic Computing München

• W. Brockmann,
• N. Rosemann

Institute of Computer Science Computer Engineering Group

• E. Maehle,
• A. El Sayed Auf,
• S. Krannich, R. Maas

K.-E. Großpietsch University of Lübeck Institute of Computer Engineering Fraunhofer Institut AIS Sankt Augustin

I I I I T T I I I I T T

ORCA – Organic Robot Control Architecture

Reflexes Motor (PWM) Motor-Controller Motors (PWM) Perception Proprioception Motor-Controllers Reflexes Behaviours Planning Gait-Pattern-Gen. Gait-Pattern-Gen. Gait-Generation /

• Selection

Gait-Pattern-Gen. Gait-Pattern-Gen. Sensor Modules decisional level functional level hardware level OCU OCUs

BCU = Basic Control Unit OCU = Organic Control Unit

Monitor Reasoner Memory

OCU-Architecture

• Monitor: anomaly detection
• Memory: short term history (learning)
• Reasoner: hard real-time determination
• f a counteraction

Variant of Observer/Controller Architecture

Characteristics of ORCA

• Application of ORCA in complex embedded automation systems
• Reduction of complexity of overall system design by structuring functionalities

and providing the system with self-x-properties

• Anomalies and uncertainties are treated in a uniform way, by encapsulating

monitoring and reconfiguration reasoning into OCU modules (inspired by immune system) -> health signal generation

• Bypassing the problem of impossibility to make complete world/environmental

and fault models by means of OCU interplay

• Incorporation of learning methods into the OCUs
• Modular structure fits object-orientated programming style

Architectural Work Package

Platform: six-legged walking machine OSCAR Features:

• 18 digital servos
• Internal servostate feedback
• Leg-(de)attachment

behavior level planning level reflex/leg behavior level

ORCA

behavior level planning level reflex/leg behavior level

ORCA

Planning level: Simulator beta

Temporarily forbidden regions Health signal

behavior level planning level reflex/leg behavior level

ORCA

Behavior level

• Movement behavior health signals are based on

monitored events, not only on sensor health signals

• Wander, avoid IR, avoid US and escape behavior OCUs

share event information

• Behavior OCU may notify neighboring OCUs to suppress

their outputs to allow execution of a behavior “self test”

• “Self tests” can include carefully executed actions where

the robot might gently touch objects

• OCUs can alter their health signals based on “self test”

results

behavior level planning level reflex/leg behavior level

ORCA

Reflex / Leg behavior level

Gap crossing (extended search reflex)

behavior level planning level reflex/leg behavior level

ORCA

Methodological Work Package

• Lower level of architecture:

– Trajectory tracking – Detection of anomalies – Compensation of anomalies  Self-optimizing, self-healing interface for higher levels, concerning – good representation of actual system state by preprocessing (anomaly compensation, virtual sensors, etc.) – the execution of commands / trajectories from higher levels

Adaptive Filters for Anomaly Compensation

• Use of adaptive filters for the treatment of

environment-induced disturbance effects

• Use of adaptive filters for the combination
• f fault monitoring and fault correction

 Compensated faults can be ignored by higher levels (self-healing)  Results of monitoring can be mapped to health signals

Self-Models for Anomaly Detection

• Machine learning approach: local

anomaly detection by self-models  Learn normal system behavior during periods of (more or less) normality

• Self-models have to be able to deal

with uncertainties / anomalies within the inputs as well as within the learning stimuli

• Self-models have to assess the

quality / trustworthiness of their own

• utput
• Challenge: handling of uncertainties

within self-models

self- model

self- model input uncertainties

• utput

uncertainties reference uncertainties

BCU

HS-based Blending of Strategies

• Aim: use anomaly information within a BCU
• Approach: blending of behaviors based on

health signals (HS)

• HS-Blending between

– a safe, but suboptimal behavior – a self-optimizing behavior

• Example:

process BCU

(safe)

input uncertainties input health signals determine strategy BCU

(self-opt)

s1 s2 s1 s2 1

Health signal Activation

OCU

Controlled Self-Optimization

process BCU OCU input uncertainties uncertain local sequential learning stimuli ill-posed learning problem  regularization required Approach: incremental regularization by the SILKE approach (System to Immunize Learning Knowledge-based Elements) So far: convergence for zero-order sTS (self-optimization)

(for specific templates)

first-order sTS (self-modelling)

Controlled Self-Optimization

• Formal framework and proof

– Analyze repetitive SILKE steps concerning the contraction property – Here, contraction corresponds to constraints on the eigenvalues of a specific matrix – Matrix depends on SILKE mask, which corresponds to the desired meta-level property





  

U u

u m u p

• p
• p

T ) ( ) ( ) ( ) (

)) ( ) ( ( ) ( ) 1 ( ) ( p T p

• p
• p
• 

      

Controlled Self-Optimization

Result: procedure for designing SILKE templates which are convergent by construction  Formal a priori guarantees of general, dynamic system properties, even though there is online self-optimization Design matrix Characteristics matrices with |eig(M)| < 1

Outlook / Final Demonstration Scenario

• Final, joint demonstration scenario based on

the OSCAR robot

• Implement and investigate the full ORCA

architecture for a rescue/monitoring scenario

• Adaptive filters for compensation of anomalies
• Self-optimizing system representation with the

SILKE / ODIL approach for anomaly detection

• Health aware behaviors and reflexes
• Health aware re-planning
• Generalization to other applications: („ORCA =

Organic Robust Control Architecture“)

References

• Rosemann, N.; Brockmann, W.: Incremental Regularization to Compensate Biased

Teachers in Incremental Learning. In: Proc. 2010 World Congress on Computational Intelligence, IEEE Press, Piscataway, 2010, 1963-1970

• Großpietsch, K.-E.; Silayeva, T.A.: Fault Monitoring for Hybrid Systems by Means of

Adaptive Filters. In: Proc. IDIMT 2010 Conf. Jindrichuv Hradec 2010, Trauner Verlag Linz, pp. 177 – 185

• Buschermöhle, A.; Hülsmann, J.; Brockmann, W.: A Generic Concept to Increase the

Robustness of Embedded Systems by Trust Management. Accepted for: 2010 IEEE Conference on Systems, Man, and Cybernetics - SMC2010, Istanbul, 10.-13.10.2010

• Jakimovski, B.; Maehle, E.: In situ self-reconfiguration of hexapod robot OSCAR using

biologically inspired approaches. Climbing and Walking Robots by Behnam Miripour (Ed.), INTECH, ISBN: 978-953-307-030-8, 311-332, 2010

• Jakimovski, B.; Meyer, B.; Maehle, E.: Firefly flashing synchronization as inspiration for

self-synchronization of walking robot gait patterns using a decentralized robot control

• architecture. Architecture of Computing Systems - ARCS 2010, 23rd International

Conference, pp. 61-72, Hannover, Germany, 2010

• Jakimovski B.; Meyer B.; Maehle E.: Design ideas and development of a reconfigurable

robot OSCAR-X. 13th International Conference on Climbing and Walking Robots and the Support Technologies for Mobile Machines - CLAWAR, 31 August – 03 September 2010, 391-398, Nagoya, Japan, 2010

• Dissertation A. El Sayed Auf: Eine Organic Computing basierte Steuerung für einen

hexapoden Laufroboter unter dem Aspekt reaktiver Zuverlässigkeit und Robustheit, Lübeck, Aug. 2010