Approaching the Schedule Execution Problem with O OSCAR Amedeo by dmpe

VIEWS: 11 PAGES: 39

									         Approaching the Schedule Execution
              Problem with O-OSCAR


              Amedeo Cesta, Gabriella Cortellessa, Angelo Oddi,
                  Nicola Policella and Riccardo Rasconi

                             Planning and Scheduling Team [PST]*
                                ISTC-CNR, Italian National Research Council
                                     Viale Marx 15, I-00137 Rome, Italy
                                    {cesta, corte, oddi, policella,rasconi}@ip.rm.cnr.it
                                                   http://pst.ip.rm.cnr.it




*Cortellessa, Policella and Rasconi are also affiliated with Università degli Studi di Roma “La Sapienza”

                                                                                                            ISTC
                       What is ISTC-CNR?
• Italian National Research Council (CNR)
   – Main governamental institution for research
   – Several Institute over the country cover several research areas
   – Re-organized in 2002 in 100 Institutes


• New CNR Structure (Feb. 2002)
   – Institute for Cognitive Science and Technology (ISTC-CNR)
        • Recognized active role both in developing new technology and in
          studying the social/cognitive aspects of technology
        • Inherited activities of IP-CNR (Institute of Psychology ) that have had
          an active AI research on several topics
        • Additional technological groups joined in (e.g., ontology, semantic
          Web, Intelligent Interfaces)



                                                                                ISTC
         Planning and Scheduling at ISTC-CNR
•   Different Aspects of research and development
     – Planning
          • What to do for achieving a goal


     – Scheduling
          • When to do actions to have enough resource for them

     – Interactive Problem Solving
          • How to include the user in the loop


     – Space Applications
          • For several reasons is a peculiar expertise in the group

•   Collaborations:
     – Univ. Roma “La Sapienza”
     – Robotics Institute, Carnegie Mellon University, Pittsburgh, USA
     – Within PLANET with several Research Centers over Europe

                                                                         ISTC
                      Present Projects
• ASI (Italian Space Agency) -- Basic research funds
   – SACSO: SAfety Critical Software for robotics (JERRY)
   – ARISCOM: Contraint-Based Continuous Planning (O-OSCAR and
     beyond)

• MIUR (Italian Ministry for Education, University and Reserch)
   – Multi-agent systems with software and robotic agents for “socially
     relevant” applications

• ESA (European Space Agency)
    – MEXAR Interactive Support for Mission Planning in MARS
      EXPRESS
    – Ended in July 2002




                                                                     ISTC
        Multi-Agent Systems for Health Care Facilities and
                     Domestic Environments
To be developed over 3 years with “low-cost” technology
Specifications:
•Fixed and mobile (heterogeneous) robotic components, with specialized roles
•Agents:
      Tasks:
      •Facility administration ß cooperation with personnel ß integrated perception & learning
      •Human assistance ß human/robot interface ß multi-media/voice interaction etc…

      Specific behaviours:
      •Follow, Find, Transport, Manipulate, Guide, Surveillance

•Planning with variable (de)centralization
•Agent level scheduling
•System has to be remotely controllable (Internet)
Project is currently in its “initial” phase: official start January 2003
                                                                                          ISTC
    Aim of our work : a complete approach to
            Plan/Schedule Life-cycle

                               Problem
                               Solving        Ex
                   lem                           ecu
                ob ation
              Pr ific                        Mo       tio
                                                         n
                                               nit
               ec                                 ori
             Sp                                      ng




                           User in Control


O-OSCAR can be used as a Decide, Monitor & Control tool
    for strategic coordination of AI agents behavior
   (e.g., flexible support to rescue domain planning)
                                                             ISTC
                      Outline


• Links with the Rescue Domain

• Description of O-OSCAR (Object-Oriented Scheduling
  Architecture)

• Schedule execution in O-OSCAR

• Conclusions




                                                   ISTC
                The Rescue Domain

• Buildings (house, fire station, hospital, police office)
   – May collapse, obstructing a road and/or burying people
• Streets
• People
   – May remain involved in some accidents
• Cars
   – May suffer breakdowns or find obstructions
• Fire brigade, ambulance, police
   – Same as cars, plus: fire brigade may run out of water and/or
      hose length may not be sufficient
   – Represent the world resources to face catastrophic events



                                                                    ISTC
                            Goals

• Environment may change, forcing reconsideration and revision
  of pre-established plan. Examples:
   – Assistance could be delayed if the police car, ambulance or
      fire brigade find obstructed streets
   – Time needed for mission completion could be unpredictable
   – etc…

• Goal 1: to produce a coordinated behavior where requests are
  served in a timely and cost-effective manner
• Goal 2: to ensure continuity maintenance, i.e., keeping the
  newly found solution as close as possible to the previous one.




                                                                   ISTC
                        Goals


Problems              events
            θ                        θ’

                                       Goal 1: resource allocation

solutions
            αθ                      αθ’

                 Goal 2: continuity maintenance




                                                            ISTC
              O-OSCAR

An Object-Oriented SCheduling ARchitecture




                                             ISTC
                    O-OSCAR


• Based on a constraint-based representation
   – Domain formalised as a CSP (Constraint
     Satisfaction Problem)



• Uses of powerful Profile Based Algorithms




                                               ISTC
    Problem Solving with CSP
(Constraint-Satisfaction Techniques)
                      Search
                      Decision

     CSP
    Problem
    Definition

                      Constraint
                      Data-Base




                     Propagation
                       Rules


                                       ISTC
            O-OSCAR’s Structure

• Constraint Data Base
   – Schedule Representation and Management

• Problem Solver

• Interaction Module

• Execution Monitor



                                              ISTC
             Constraint Data Base

• A constraint engine based on quantitative temporal
  constraints and resource constraints

• Within a given domain, allows the definition of a wide
  range of problems in terms of
   – activities to be executed
   – resource requirements
   – existing temporal constraints among the activities




                                                       ISTC
                  Constraint Data Base
The Constraint Data Base
represents the situation of
interest

                                          Real World




         Constraint Data Base              Domain
                                          & Problem
                       Current Solution   Description
                                          Languages
          Temporal       Resource
         Constraints    Constraints



                                                        ISTC
                Constraint Data Base
               Domain Representation Layer
               -- Activity          -- Resource
               -- Constraint        -- Decision


           Constraint Representation Layer

          Temporal Constraints   Resource Constraints


                                                        Renewable &
Dynamic
                                                        Consumable
 STP
           TEMPORAL NETWORK        RESOURCE PROFILES     Resources


                                                              ISTC
        Schedule Representation and
               Management

• It is responsible for the maintenance of the solution
  which is incrementally refined while being computed

• Uses the Constraint Data Base, which makes sure that
  all constraints representing the problem domain are
  satisfied at all times




                                                          ISTC
                Interaction Module



• Provides the user with constantly updated information
  about the schedule status

• Provides a number of straightforward functionalities
  devoted to user-system interaction




                                                          ISTC
              Interaction Module

User Interaction Module



                                 Real World




 Constraint Data Base             Domain
                                 & Problem
              Current Solution   Description
                                 Languages
 Temporal       Resource
Constraints    Constraints



                                               ISTC
             User supports for problem definition
                                 An environment for visualizing
Problem                              and refining a problem
definition
facilities
                                                Graphical editor
                                                  for on-line
                                                   problem
                                                 modification




                                                 Information on
                                                 current solution



                                                         ISTC
           Specialised problem visualisation
                                   A Resource Constrained
                                  Project Scheduling Problem




Coded in           Nodes are activities and edges are
  Java             temporal precedence constraints

                                                         ISTC
                    Problem Solver
Different active services
                                      Planning / Scheduling
are implemented on top of
                                      System
the constraint representation

    Problem                            Real World
     Solver



      Constraint Data Base              Domain
                                       & Problem
                   Current Solution    Description
                                       Languages
      Temporal       Resource
     Constraints    Constraints



                                                          ISTC
          “Profile-based” Algorithms

• Based on a general procedure which:
   – maintains a temporal consistent solution

   – attempts to resolve any encountered resource
     conflict by imposing new precedence constraints
     between competing activities

   – flattens any contention peaks possibly present in
     the resource usage profiles.



                                                         ISTC
                             “Profile-based” Algorithms

                                   max capacity


            Activity B
       Activity A




         leveling                  max capacity
                                                            Meta-CSP
                                                            (MCSs are
                                                             conflicts)


                Activity B
Activity A

                                                          Ground-CSP
                                                           (temporal)




                                                             ISTC
User support during/after problem solving
                            Graphical Representation
                                 of the problem




                                 Resource Profiles



                                         Solution
                                        Gantt Chart




                                               ISTC
Different perspectives on the same solution




                                      Resource
                                        level




                 Resource conflict
                                          ISTC
                   Plan Execution

• Dispatching of activities according to temporal order

• Update of current schedule representation according to
  execution feedback

• Possibility of reactively revising the solution according
  to automated rules or manual modifications




                                                          ISTC
             The Execution Monitor


• Monitor schedule’s execution (execution phase) by
  implementing a basic cycle in which every activity is
  dispatched to be executed at proper times

• Performs a dynamic rescheduling on occurrence of
  unexpected exogenous events




                                                          ISTC
        The Execution Monitor
                                 Execution Monitoring
                                 System


                  Execution      Real World
                   Monitor



 Constraint Data Base
              Current Solution

 Temporal       Resource
Constraints    Constraints



                                                    ISTC
            The Execution Monitor

• Possible disturbing events:
   – unexpected delays of activity start time
   – unexpected changes in activity duration
   – unexpected decays in resource performance

• Two main approaches are generally followed:
   – predictive approach
   – reactive approach (used in O-OSCAR’s Execution
     Monitor)


                                                      ISTC
               The Execution Phase

• O-OSCAR modules involved:
  – Constraint Data Base
     • whose services are also needed in this phase

  – Interaction Module
     • used to provide the user with visual information on schedule
       evolution

  – Execution Monitor
     • developed ad-hoc




                                                                 ISTC
        Basic Plan Execution Structure

• Dispatching activities according to temporal order


• Update of current schedule representation according to
  execution feedback


• Possibility of reactively revising the solution according
  to automated rules or manual modifications




                                                              ISTC
           The basic Execution algorithm

• The Execution Monitor periodically checks for inconsistencies
  between sensed real values and expected values.

• If differences are detected
   – the Constraint Data Base gets immediately updated to reflect the
     new situation (propagation)
   – if propagation yields a temporal inconsistent situation
        • failure is declared and execution terminates
   – otherwise
      • execution may continue, yet the schedule must undergo
        further inspection in order to highlight the presence of possible
        resource contention peaks. If any are found, the solution
        needs to be fixed before execution can continue

                                                                      ISTC
                    Execution Viewer
        Time = 10



Executing
Activities:
2, 6, 15, 18




Delayed
Activities:
1, 21



                                       ISTC
             Evolution of the schedule
                                         Time = 43
Executed
Activities




                  Additional delays
                                                 ISTC
     Using O-OSCAR in the Rescue Domain


To satisfactorily tackle the Rescue Domain Problem,
we choose the approach of integrating different
complementary tasks, each requiring specific attention:

 • Specification of the problem at hand, performed by the user
 • Interactive problem solving procedure, which leads to the creation
   of the initial recovery schedule
 • Schedule execution, which is continuously monitored and
   controlled by the user in order to ensure satisfaction of the goals




                                                                    ISTC
A Constraint-based Software Approach to
    Activity Planning and Scheduling

               Domain Description
                   Language


                Constraint- Based
                 Representation
              (Constraint Data- B a s e )


                                                                 End-to-end
                                                                 application




Solution Methods                        Interaction with Users




                                                                        ISTC
                        Conclusions

• Software architectures for planning, scheduling & execution
  monitoring

• Component technology for achieving control in planning Rescue
  Domain problem operations

• Model-based knowledge representation supported by constraint-
  based techniques

• Problem solving techniques based on constraint reasoning +
  execution control mechanisms




                                                                ISTC

								
To top