Plan Representation and Reasoning with Description Logics
Representing Planning Knowledge in Description Logics: Overview • Action taxonomies in CLASP – extended language to represent action networks • Plan taxonomies in SUDO-PLANNER – plan subsumption of partially ordered plans • Goal taxonomies in EXPECT – expressive representations of goals and their parameters These systems can exploit the descriptions of all the objects in the domain ( domain knowledge ) in order to reason about action, goal, and plan descriptions
Defining Actions, States and Plans in CLASP in a Telephony Domain (DEF INE-PLAN Po ts - P lan (DEF INE-CO NCEPT Sys tem-Ac t (AND P lan (AND Ac t i on (ALL PLAN-EXPRESSION (ALL ACTOR Sys tem-Agen t ) ) ) (SEQUENCE (SUBPLAN (DEF INE-CO NCEPT Connec t -D i a l t one -Ac t O r i g i nat e -And -D ia l -P lan ) (AND Sys tem-Ac t (TEST (ALL PRECONDIT I O N (Ca l l ee - On-Hook -St a t e (AND O f f -Hook - Sta te (SUBPLAN Te rm ina te - P lan ) ) I d l e -S ta te ) ) (Ca l l ee -O f f -Hook -S ta t e (A l l Add - L IST D ia l t o ne -S ta te ) (SEQUENCE (ALL DELETE-L IST I d l e -S ta te Non -Term ina te -Ac t (ALL GOAL Ca l l e r - On-Hook -Ac t (AND O f f -Hook - Sta te D i sconnec t Ac t ) ) ) ) ) ) ) D ia l t one - Sta te ) ) ) ) (DEF INE-PLAN Or i g i n a te -And -D i a l -P lan (DEF INE-CONCEPT Ca l l ee -O f f - Hook -S ta te (AND (PR IMIT I VE S ta te ) ) P lan (DEF INE-CONCEPT Ca l l ee -On-Hook -S ta te (ALL PLAN-EXPRESSION (PR IMIT I VE S ta te ) ) (SEQUENCE (DEF INE-CO NCEPT Ca l l ee -O f f - Ca l l e r -On - Sta te Ca l l e r -O f f -Hook -Ac t (AND Ca l l ee -O f f -Hook -S ta te Connec t - Dia l t one -Ac t Ca l l e r -On -Hook -S tat e ) ) D ia l -D ig i t s -Ac t ) ) ) )
Plan Representation and Subsumption in SUDO-PLANNER • Plan is described as a set of action types associated with identifiers – [(surgery, id1) (CABG, id2)] • Plan is simplified if action subsumption and same id – [(surgery, id1) (CABG, id1)] -> [(surgery, id1)] • Plan subsumption – Action network viewed as bipartite graph matching a4 a2 a5 a5 a1 a1 R P S Q a3 a4 a6 a6 a2 a3
Matching Goals in EXPECT • Desired goals and available capabilities are automatically translated to LOOM concepts • Classifier is used to find most specific method OBJ capability that subsumes the posted goal cargo Method capability: move OBJ WITH (move cargo vehicle (OBJ ( ins t -o f ca rgo) ) move WITH (W ITH ( ins t -o f a i rcra f t ) ) ) aircraft Goal: OBJ (move OBJ cargo move cargo (OBJ ( ins t -o f ca rgo ) ) WITH move (WITH C-140 ) ) WITH C-140 truck Self-organizing method taxonomy
Reactive Systems Yolanda Gil CS 541, Fall 2003 (Thanks to Karen Myers from SRI International)
Summary • Control systems – Networks of “variables” (arcs) and “functions” (nodes) • Reactive Action Packages (RAPs) – Networks of “conditions” and “tasks” • Task Control Architecture (TCA) – Network arranged according to “vertical capabilities” • Procedural Reasoning System (PRS) – Integrates planning, BDI, and reactive techniques • Anytime algorithms – When time is short, managing what you think about • Learning and uncertainty reasoning
PRS Interpreter Execution Cycle 1. New information New Facts & Tasks arrives that 2 (overpressurized fuel-tank) 7 updates facts (ACHIEVE (position ox-valve closed)) Act Library and/or tasks 1 2. Acts are triggered by new facts or ACT2 Cue: tasks (TEST (overpressurized tank.1)) Act Execution Facts ACT1 Cue: 3. A triggered Act is (ACHIEVE (position valve.1 closed)) & 6 intended External Tasks World 4. An intended Act is selected 5. That intention is activated 8 5 6. An action is Goal2 Goal3 performed ACT8 ACT3 sleeping sleeping 3 (ACHIEVE (position ox-valve closed)) 7. New facts or ACT1 current Fact1 tasks are posted ACT2 4 normal 8. Intentions are Intention Graph updated
Distributed and Multi-Agent Planning
Issues • Who is in charge? • How distributed? • How much info is shared? • Who benefits? • What and how to communicate? • How and how much to coordinate? • Can tasks/goals/resources be negotiated? • How to handle execution dynamics?
Summary (I) 1. Task sharing – Homogeneous agents – Heterogeneous agents • Contract nets – Contactors bid – Managers bid • Market mechanisms 2. Results sharing – Blackboard architectures – Distributed constraint satisfaction – Resource sharing through auctions
Summary (II) 3. Distributed planning – Planning approaches • Cooperative plan Construction • Centralized planning for Distributed plans • Distributed planning for Centralized plans • Distributed planning for Distributed plans – Execution issues • Post-planning • Pre-planning 4. Mental state and collaboration – Joint intentions – SharedPlans 5. Coordination without communication
Mixed-Initiative Planning
Challenges • Interpreting user input – Mapping into possible operations/responses – Disambiguating requests • Intelligent search – Managing classes of solutions – Tracking constraints and previously explored solutions • Facilitating user’s cognitive task – Grounding the discussion with a specific plan • Acting on the user’s input – Flexible planning framework that can support collaboration
Recap and Summary • Dialogue issues in mixed-initiative planning: TRAINS – Interpreting user input, disambiguation – Plan representation as goals/tasks/resources/state + straw plan – Hybrid planning architecture • Integrating user guidance with a planning algorithm: PASSAT – Incorporating the user’s input into a plan generation algorithm
Recommend
More recommend
