Intelligent Systems

Software Assurance Symposium 2004

Bojan Cukic & Yan Liu, Robyn Lutz & Stacy Nelson, Chris Rouff,

Johann Schumann, Margaret Smith

July 22, 2004

“What”

• Intelligent Systems research will create “new generations of robust, fault-tolerant software for intelligent, cooperative space systems that operate largely autonomously from ground control” --NASA list of key technology areas for H & RT Advanced Space Technology, 6/04

• New technologies for V&V of Intelligent Systems

“What” (cont.)

• Technologies demonstrated at this year’s presentations:– Neural Networks– AI Planners – Support Vector Data Description algorithms– Bayesian-based safety envelopes– Autonomous contingency identification and

recovery technology– Model Checking– Hybrid formal methods

Information Systems Presentations

ScheduledPresentations

By TRL Real World Application

Verification andValidation of AdaptiveSystems

BojanCukic

5 IFCS, F-15

Bayesian Verificationand Validation toolsfor Adaptive Systems

JohannSchumann

7 F-15

Formal Approaches toSwarm Technologies

Chris Rouff 2+ ANTS

Information Systems PresentationsContingencySoftware inAutonomousSystems

Robyn Lutzand StacyNelson

2+ ARP

ModelChecking ofArtificialIntelligenceBasedPlanners

MargaretSmith

7 DS-4/ST-4JPL S/C

LyapunovStabilityAnalysis andOn-LineMonitoring

Bojan Cukic 5+ IFCS JPL’s FaultProtection engineEarth Orbiting

SSatellite

Intelligent Systems: Why ?

• Long lived missions

• Lower operations costs

• Swarms & constellations of satellites/spacecraft

• Currently used in other domains:– automotive– health– waste water management

• Intelligent Systems are here to stay!

Intelligent Systems: Why not

• Is the technology:– Scalable for usage?– Being oversold?– Just a piece of a larger puzzle?

• V&V of Intelligent Systems requires a new knowledge set: math, tools, control theory, and highly skilled software engineers.

• V&V is scrambling to catch up to new technologies for Intelligent Systems

Directions?

• Do we know yet how to design intelligent systems for verifiability? (or meaningless to lump them?)

• Is the IV&V process different for intelligent systems?

• Are we ready to demonstrate scalability on real systems?

• Should we be developing V&V standards for intelligent systems? Tied to criticality levels?

• How do we start establishing benchmarks for intelligent systems?

Verification and Validation of Adaptive Systems by Bojan Cukic

• Investigate the role of modern AI techniques (Support Vector Machines) in failure detection and identification.– Failure Detection

• Designing a fast (real-time) SVDD algorithm to detect failure conditions

– Failure Identification• Failures are identified by studying the correlation

between certain longitudinal and lateral dynamics parameters

– Validate the technology in extensive simulations

Bayesian Verification and Validation tools for Adaptive

Systems by Johann Schumann

Problems with traditional V&V methods applied to Adaptive Systems:

•Fault avoidance design testing applies to base case only

–Unanticipated failures?

– Unmodeled failures?

•Fault removal cannot test all possible configurations in advance

•Fault tolerant design does not consider all possible problems

–

Bayesian Verification and Validation tools for Adaptive

Systems by Johann Schumann

Methods for improvement:

•Improve performance estimation of the neural network (Bayesian approach)

•Use Envelope tool to answer:

– How large is the current safe envelope?

– How far is the operational point from the edge?

Formal Approaches to Swarm Technologies by Chris Rouff

• Survey formal approaches for agent-based, multi-agent and swarm-based systems for appropriate swarm-based methods• Apply most promising approaches to parts of ANTS• Evaluate methods for needed properties• Model and outline swarm-based formal method• Develop formal method for swarm-based systems• Do formal specification of ANTS using new method• Prototype support tools

Formal Approaches to Swarm Technologies An ANTS Overview - by Chris Rouff

Earth

Lagrange Point Habitat

1. Assembly & release

2. Self propelled transit

6. A messenger carriesfindings to Earthwhen needed.

4. Swarm (Fly by) Operations

5. Repeat steps 3 and 4.IR Worker

MAG Worker

X-Ray Worker Messenger

Asteroid belt

3. Long-Range Operations

Asteroid(s)

Workers

Messengers

Rulers

Workers

Workers

M. L. Rilee, EIT, S. A. Curtis, NASA/GSFC, 2001.

ANTS: Mission Concept 2020

Contingency Software in Autonomous Systems by Robyn Lutz

& Stacy Nelson

The Goal - Mitigate failures via software contingencies resulting in safer, more reliable autonomous vehicles in space and in FAA national airspace

How?

• Adding intelligent diagnostic capabilities by supporting incremental autonomy

• Responding to anomalous situations currently beyond the scope of the nominal fault protection

• Contingency planning using the SAFE (Software Adjusts Failed Equipment) method

Model Checking of Artificial Intelligence Based Planners

by Margaret Smith

• Goal: Using model checking, and specifically the SPIN model checker, retire a significant class of risks associated with the use of Artificial Intelligence (AI) Planners on Missions – Must provide tangible testing results to a mission using AI technology.– Should be possible to leverage the technique and tools throughout NASA.

• FY04 Activities:– Identify and select candidate risks– Develop and demonstrate technique for testing AI Planners/artifacts

on:• A toy problem (imaging/downlinking) – demonstrate tangible results with

an abstracted clock/timeline• A real problem (DS4/ST4 Champollion Mission) – demonstrate, using

DS4 AI input models, that Spin can determine if an AI input model permits the AI planner to select ‘bad plans’.

Lyapunov Stability Analysis and On-Line Monitoring by Bojan Cukic

The Problem:

•Issues with Adaptive Systems: uncertainty/newness

•Need Understanding of self stabilization analysis techniques suitable for adaptive system verification

•Need to investigate effective means to determine the stability and convergence properties of the learner in real-time

The Approach:

•Online Monitoring

•Confidence Evaluation

Lyapunov Stability Analysis and On-Line Monitoring by Bojan Cukic

Relevance to NASA:

• Artificial Neural Networks are increasingly important in flight control and navigation

• Autonomy and adaptability are important features in many NASA projects

• The theory is applicable to future agent-based applications