Www.CrewAssistant.com 1 The MECA Project Reasoning Agents on Mars Leo Breebaart (S&T) 12 October...

www.CrewAssistant.com 1

The MECA ProjectReasoning Agents on Mars

Leo Breebaart (S&T)

12 October 2006


The MECA Consortium

TNO Human Factors• Human behaviour and performance in technical high-demand environments;

methods to attune the environment to (momentary) human capacities.

S&T bv • Software development company with specific expertise on creating system

health management applications.

OK Systems• Software development company focusing on technology areas of AI, user

interfaces, databases and web-based systems (specially scheduling systems).

EADS-ST• Technologies, development, production and utilization of manned and

unmanned space missions, including experiments, space transportation systems, propulsion systems and support of these systems concerning operations, maintenance and mission handling.


Project Goal

The MECA Project aims to:

• Provide support to and increase autonomy of astronauts while:– Executing complex tasks– In a hostile and large unknown environment– Possibly disconnected from Mission Control


Objective & Vision

Objective: support mission goals (without injury or loss of life) by• empowering the cognitive capacities of human-machine teams

during planetary exploration missions. • in order to cope autonomously with unexpected, complex and

potentially hazardous situations.

Vision: crew support that • acts in a ubiquitous computing environment • as “electronic partner”, helping the crew

– to assess the situation, – to determine a suitable course of actions to solve a problem, – to safeguard the astronaut from failures.


MECA and Agents

Where does agent technology fit in?

• A MECA System can be considered as an instance of a group of software agents.

• The MECA Architecture will make use of the outcome of agent technology research.


This presentation

• Phase 1 (2005-2006)– “MECA 2017”– RB: Requirements Baseline

• Phase 2 (2006-2007)– “MECA 2007”– Demonstrator Prototype– Refined RB


Why the need for support?

• Current astronaut support rather antiquated.

• Long-distance manned explorations impose new challenges.


Background


State of the Art

LAPAP SCOPE

Our legacy:…

But MECA should extend this….

Operational Procedures Hardcopy +


MECA design process

Background • Operability-based design• Anticipate new Intelligent Interfaces• Anticipate new HFE standard

Approach• Human-Operation centered• Enabling Technology focus• Iterative process (specify-test-refine

cycles)• From abstract to detailed specifications• Sound theoretical and empirical foundation

10 -- 25 yrs


Example Scenario

Human and System Health Management


EVA 2 astronauts 2km from habitat, sample collection for scientific experiment.

Habitat

= MECA unit


MECA indicates a problem with the temperature regulation of one of the space suits.

Habitat


Astronaut, MECA and spacesuit collaborate on finding the cause of the failure.

Habitat

• MECA interrogates spacesuit to obtain more parameters for diagnosis.


The heater of the space suit is damaged and cannot be repaired locally.

Habitat


MECA simulates the consequences of the broken heater.

Habitat

• MECA predicts that astronaut has to be brought to habitat and has a risk of fainting.


Astronaut's MECA Unit and Habitat MECA Unit reschedule and plan safe return.

Habitat

• Habitat is preparing for treatment of hypothermic astronaut (preparing and activating resources)• Astronaut MECA asks for help of transporting astronaut (since he is likely to faint)


MECA Rovers respond to call for help, MECA habitat chooses rover

Habitat

• MECA habitat chooses rover that has enough power/resources to pick astronaut up and transport to habitat (fuel, location, speed, pressurized or unpressurized rover etc.).


MECA communicates adjusted schedule to astronaut(s) in the right manner (keeping in mind cognitive task load).

Habitat

• Hypothermic astronaut has a high task load due to high stress levels, MECA Unit shall adapt communication according to task load and affective state.• The astronaut accompanying him can be given information in a different manner.


Astronaut faints before rover arrives. MECA communicates fainting of astronaut.

Habitat


Astronaut has fainted earlier than predicted, MECA rover has to find a way to pick up astronaut (sensemaking).

Habitat


MECA of fainted astronaut and other astronaut collaborate to get fainted astronaut in rover

Habitat


Hypothermic astronaut is transported to habitat by rover.

Habitat


Hypothermic astronaut is in habitat being treated.

Habitat


Iterative Requirements Analysis

MECA Requirements

Refine HMC Performance

Simulation-Based Evaluation

Envisioned Technology

Human Factors Knowledge

OperationalDemands

PrototypeCurrent

Technology

System DesignReview

Comments

ScientificDiscourse


Operational Demands

MECA shall take account of:

• the high-level operation goals – e.g., safe return to earth

• the environment – e.g., radiation and social monotony

• task performance – e.g., people will get seriously ill


Human Factors Demands

• Cognitive Task Load

• Situation Awareness and Sense Making

• Diversity of Cognitive Capacities

• Trust and Emotion

• Collaboration

• Crew Resource Management

• Decision Making


Envisioned Technology (1)

habitatMEVhabitatfacility

MCC

Orbiter

crewrover

= meca unit

An infrastructure will be available for automatic distribution of data, software and reference documents.


Envisioned Technology (2)

• MECA shall make use of this infrastructure and can cope with possible failures

• Continuous analysis and extrapolation of emerging technologies, e.g. – multi-agent systems– automatic planning and scheduling– model-based health management

• Technical requirements, such as– maturity– graceful degradation– maintainability– fault tolerance


Requirements Baseline

Generic task level requirements

• Implement key Human Factor knowledge.• Enhance autonomy of individual actors and groups of

actors. • Support collaboration among the different actors. • Hardware and software systems must be highly reliable. • Manage environmental information.


Current Requirements

• Different types of requirements:– task level, – functional, – user interface, – technical interface, – operational and – technical requirements.

• All requirements are linked to use cases.


Use Case Template

ID 78

Title Hypothermic astronaut

Level Level 0

Goal Treat hypothermic astronaut that is on EVA…

Actor Personal MECA, MECA habitat, astronaut in habitat, rover..

Pre-condition Astronaut on EVA is hypothermic

Post-condition Hypothermic astronaut is in medical facility being treated

Frequency Not frequent

Main success scenario

Personal MECA detects hypothermia and communicates to hypothermic astronaut …

Alternative scenario

.. , Doctor is not in habitat, MECA will ask astronaut_1 to prepare…

Comment Derived from RefDoc3…


Outline of Functional Requirements

Process MECA function

Information Gathering detect needs for operations and training

Goal Setting select and prioritize goals for operations and training

Plan Generation or Selection generate plans, or select pre-generated plans and procedures, for operations and training

Plan Evaluation evaluate operational and training plans

Prepare for Execution prepare the resources for executing operational and training plans

Execution execute operational and training plans

Processing Evaluation of Results

evaluate execution results for operational and/or training purposes


Incremental and Iterative Prototyping

• Human-, task- and context-driven design and evaluation.

• Both MECA and the humans will show mutual adaptive behaviour, which effects should be well tested with realistic scenarios.

• Prototyping, simulation and testing is therefore essential to establish a sound and coherent set of requirements.

• A game-based simulation environment can provide an effective platform for testing the human-machine collaboration (e.g. the Unreal Tournament game-engine) in combination with other simulators.


Evaluation Criteria

• Long-term human in the loop effects• Standard usability measures

– effectiveness – efficiency – satisfaction – learnability

• Human experience measures, such as – situation awareness (perception, comprehension and

projection)– trust (persistence and behavioural competence,

servitude, and the understanding of the machine)– emotion (arousal and valence)


Break?

• Break!


Background

habitatMEVhabitatfacility

MCC

Orbiter

crewrover

= meca unit

Reasoning will involve complex system behaviorand scarce resources


Background

• Time is a scarce resource• Control of a wide array of tasks and experiments• Control of each task and experiment is complex

– Nominal -- Correct sequence of steps– Off-nominal -- Detect, isolate, and compensate failures

• Summary: – Optimize crew time by supporting control activities

• Check plan execution• Support control of complex equipment especially in the case of

malfunctions• Resource usage


Human decision making

Process UnderControl

Understand Act

•Situational awareness, e.g.:• Warning: At current rates, resources will be depleted within two days…

MECA Unit

• Must reduce load by 10% but must also generate more oxygen…• I lowered the consumption, but I still don’t see any change..

Complex, uncertain, and dangerous world

plan


Traditional support

Operational procedure:

verify

action

Next action

okay

Notokay

off-nominal proceduresguided bymission control

Based on operational procedure and Mission Control


Operational procedures

Nominal control

Off-nominal:Fault detection, isolation,repair

Operational procedures:•Huge pile of papers•No feedback of payload


The 2017 MECA system

• The 2017 MECA system helps the astronaut to make the right decisions in situations that :– Are novel, or near-novel, e.g., because equipment is

failing– Are complex, e.g., effects of the decision are hard to

predict• Intricate interactions between processes, systems,

components, …• The effects are noticeable only late in time

– Require human--human and human--machine cooperation


Automated support


Understand Act

MECAUnit

plan

•Improve situational awareness by interpretation measurement data, in particular: fault diagnosis

•Determine alternative plan (e.g., repair, reconfiguration)

•High level plan --> control action•Monitor plan execution


Automated support

Validate success of plan step

Automatic fault detection and isolationAdequate repair procedurespresentation of background info

Automatic (re)plan


Automated support

• Operational procedure is generalized by Actions of a Plan– Plan describes pre- and postconditions– Postconditions used for verification plan step and diagnosis

• Automated diagnosis

• Automated reconfiguration (repair, or redundancy management)

pre post pre post pre post

Supply pressure to fuel and oxidizer

Pb = high


Layered Reasoning


2017 MECA: Collaboration


Understand Act

PUC

Understand Act

I need your capabilities to repair the equipment

Resource level fromteam mate is enough tocomplete task…


Plan-based Architecture

PUCstate

PUC

Health mode &

Resource

Healthestimation

Resourcemonitor

Plan representation

goal

subgoal subgoal

task task

Plan/schedule

Reconfiguresystem

PossibleActions,Safety

constraints,utility

Derivecapabilities

Goals


MECA Unit Functional Decomposition

MMI

procedureexecutor

(PE)

what-ifsimulation &

rehearsal (PR)

executionmonitoring (SM)

sense-making

reconfiguration(CO)

physical equipment/facility interface (DA)

planning &scheduling

(PS)

derivecapabilities

(SA)

health &status

monitoring(FDIR)

inter-faceto

otherMECAunits

CTL

resourcemanager

healthmonitoring

plan

model

status+

history

collaboration

(simulated) PUC

other MECAunits

other MECAunits

usecases


Software Architecture


MECA 2017 is not MECA 2007!

For example MECA 2017 (‘Real’ MECA) MECA 2007 (Demonstrator)

Software Technologies Make generic choices: ontologies, agents, intelligent reasoning support.

Choose specific instantiations: OWL/RDF, Jade, Uptime.

Hardware Platforms Wearable / ubiquitous / brain-jack interface...

Tablet PC / Laptop.

Autonomy of components in Process under Control

Components can be independently autonomous (i.e. non-MECA intelligence).

No (interface to) 3rd-party intelligence, all autonomy implemented / controlled by MECA.

Reasoning Optimal combination of proactive and reactive.

Combination of proactive and reactive.

Unit complexity Completely hierarchical systems-of-systems nature of MECA Units.

Only simple aggregation and limited levels of containment for MECA Units.

Cooperation Adaptable to dynamic leadership, based on model on teamwork.

Limited models of teamwork.

Cognitive task load Aware of task load based on physiological sensors , imposed loads, and actual astronaut reactions.

Limit awareness of task load based on limited indicators.


Challenge Areas

• Data Architecture: how to ensure worldviews of different levels of MECA Units remain compatible? How is the worldview represented?

• (Unit) Collaboration: how to synchronize, authenticate, de-conflict, negotiate work share, re-plan when two MECA Units ‘meet up’?

• Good interfaces will be key!


Ontologies and reasoning

• Use ontologies to describe and reason about:– Design of MECA– Capabilities and Domains

• Equipment (PUC, Resources)• Environment• Tasks• Time• Communications• etc...

• Formal ontology representation allows:– automated validation of instance data– generation of documentation, templates, code– fit in with future “semantic web” approaches


Ontology example

@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .@prefix meca: <http:///esa.int/meca#> .

# OWL Classes:

meca:Unit a owl:Class ;

# Owl Properties:

meca:isPersonalUnit a owl:DatatypeProperty; rdfs:domain meca:Unit; rdfs:range xsd:boolean .

meca:autonomyMode a owl:ObjectProperty; rdfs:domain meca:Unit .

meca:synchronisationPolicy a owl:ObjectProperty; rdfs:domain meca:Unit.


Planning & Scheduling - Design Concepts

• Mission objectives are predefined, and default general plans are arranged before mission begins.

• Short term goals and plans will be decided and controlled autonomously by crew during the mission, adapting to changing circumstances.

• Continuous planning, scheduling and rescheduling of the tasks assigned to a heterogeneous team is a complex process that consumes time and determines the chances of success.

• MECA should assist in the generation of plans, reduce the overload of crewmembers, check constraints and conditions, and optimize the usage of time and resources.

• Crewmembers should have quick and easy access to all information related with plans, be able to evaluate alternatives and make changes.


Planning & Scheduling - External Interfaces


Planning & Scheduling - Internal Components

• Rule system: editable by users, enabling explanation of decisions• Inference Engine: using Rules to generate and optimize Plans and Schedules• Representation: in a format that facilitates distribution & collaboration


Demonstrator – First Design Iteration

• Decompose scenario into Sequence Diagrams (bring time dimension into play, identify actors)

• Decompose scenario into Activity / State Diagrams

• Illustrate actor timelines with storyboards

• Identify order and depth to which storyboards are to be implemented


Demonstrator – Software Proposals

• Application Framework: Use EADS Java Framework for personal MECA Unit GUIs.

• Use Jade for implementing agent-aspects of Units themselves.

• Use Uptime for Model-Based Programming: autonomous planning, reconfiguration and fault diagnosis.

• Use RDF/XML+OWL for Knowledgebase.• Use Students’ models for PuC Components• ...


Questions?

Photo: E

SA

/Aurora

Www.CrewAssistant.com 1 The MECA Project Reasoning Agents on Mars Leo Breebaart (S&T) 12 October...

Documents

Transcript of Www.CrewAssistant.com 1 The MECA Project Reasoning Agents on Mars Leo Breebaart (S&T) 12 October...