technical coordinating unit on robot planning
DESCRIPTION
Technical Coordinating Unit on Robot Planning. TCU Participants. Aegean University, Vangelis Kourakos Brescia University, A.Gerevini Freiburg University, B.Nebel Genova University, E. Giunchiglia GMD, J.Hertzberg IRST, P.Traverso LAAS, M.Ghallab Linköping University, P.Doherty - PowerPoint PPT PresentationTRANSCRIPT
http://www.planet-noe.org
European Networkof Excellence in
AI Planning
Technical Coordinating Unit on
Robot Planning
Robot Planning TCU 2
TCU Participants• Aegean University, Vangelis Kourakos• Brescia University, A.Gerevini• Freiburg University, B.Nebel• Genova University, E. Giunchiglia• GMD, J.Hertzberg• IRST, P.Traverso• LAAS, M.Ghallab• Linköping University, P.Doherty• Lund University, J. Malec• Madrid Carlos III University, D.Borrajo• Munich University, M.Beetz• NCSR Demokritos, C. Spyropoulos, • ONERA, P.Fabiani• Orebro University, A.Saffiotti• Rovira i Virgili University, M.A. Garcia• Siemens Corp. Techno., W.Feiten• Ulm University, S. Biundo, B.Schattenberg
Robot Planning TCU 3
Roadmap
• Problems and challenges in Robot Planning• State of the art• Technology maturity • What needs to be done• Research at TCU Nodes
Robot Planning TCU 4
Robot planning
A computational activity that relies on:– Models of environment and robot – Specification of required goals and utilities– Online input from sensors and communication channels
to perform prediction and choices for achieving goals and utilities
• Robotics without planning: handcoded environment structure, strategy and goals into purely reactive control
• Specifics of planning in robotics: – heterogeneous partial models and state information, – direct integration of planning to acting and sensing
Robot Planning TCU 5
Purpose of planning in robotics
Improve the performance of robots for carrying out – a set of tasks in – a class of environments
Performance criteria:• Robustness of the behavior of robots with respect to:
– Variety of environments– Diversity of tasks
• Quality and cost of:– Environment modeling– Modification and verification of robot controllers– Interaction with and among robots
Robot Planning TCU 6
Forms and components of robot planning
• Path and motion planning: Geometric path and control trajectory along that path
• Perception planning: – Which information is needed, and when– Where to look for it– Which sensors are most adequate, and how to use them
• Navigation planning: Strategy using localization primitives, visual servoing and other
sensory-controlled motion primitives, for reaching a goal, exploring an area
Robot Planning TCU 7
• Manipulation planning: Similar strategy of sensory-motor primitives using forces and
touch (haptics), vision, etc, to handle objects and assemblies
• Task planning: Classical set of state-transition operators, with
Time and resource allocation, dynamic environment, uncertainty
• Communication planning: In multi-robots cooperation, Man-machine interaction
How to query needed information, which feedback is required
Forms and components of robot planning
Robot Planning TCU 8
PLANET is not active in all these areas
Robot Planning TCU has to be aware of them and their common features:– Uncertainty
– On-line constraints
– Dynamic environments and feedback loops
– Multi-agency
Robot Planning TCU has to build on top of them
Forms and components of robot planning
Robot Planning TCU 9
Challenges and requirements
Integration of sensory-motor capabilities
with deliberative, goal-oriented capabilities
Integrate planning to acting
Integrate heterogeneous representations: Space, time, kinematics and dynamics, physics of sensors,
uncertainty, logical properties, various constraints, including
computational
Integrate various forms of planning
Integrate planning and learning
to extend sensory-motor controllers
Robot Planning TCU 10
State of the art
Available material
• Good surveys and reference works in the literature on basic robot planning components, e.g.– Path and motion planning
– Task planning
• Roadmap surveys current work on– Integration of planning, acting and sensing (Ulm Univ.)
– Uncertainty in robot planning (ONERA)
– Integration of planning and learning (Madrid)
Robot Planning TCU 11
Maturity
Mainly of planning components :• Path and motion planning : a well mature technology
Techniques: computationnal geometry and probabilistic algorithms
QuickTime™ et un décompresseur sont requis pour visualiser
cette image.
QuickTime™ et un décompresseur sont requis pour visualiser
cette image.
Robot Planning TCU 12
Maturity
• Task planning : a wealth of algorithms in the classical frameworkTechniques: search, heuristics, disjunctive refinement, CSPs
• Perception planning: some focused problems well solved, e.g. viewpoint selectionTechniques: mathematical programming
• High level reactive controllers: preprogrammed goal-directed behaviors– PRS, Propice, RAP, SRCs and similar systems well
advanced and integrated to sensory-motoric level– Most laboratory robots run on them
Robot Planning TCU 13
Possible technology transfer
• Within roboticsThere are other limiting factors for industrial deployment, e.g.
sensory-motoric functions, reliability, security and cost– Reactive controllers: the easy first step– Special purpose navigation and perception planning
capabilities
• In other domains– Path and motion planning: in CAD, Animation and Graphics– Tasks planning with time and resources: in manufacturing,
process planning, workflow management, network management
– Reactive controllers: in transportation systems– Integrated planning systems in well structured domains:
autonomous spacecraft– Perception planning in surveillance systems
Robot Planning TCU 14
What needs to be done
• Research on integration problems– Planning and acting
• Reaction / deliberation architecture
• Planning depth and on-line constraints
• Execution models
• Fault detection, diagnosis, recovery
– Heterogeneous representation
– Heterogeneous planning techniques
– Planning with information gathering and sensing• Access to world state
• Planning (with) action sensing
– Planning and learning
Robot Planning TCU 15
Typical projects
• Cooperating service robots
Transportation, surveillance, cleaning, search and gathering of
object, e.g., office assistants
• Surveillance and monitoring of the traffic network
• Exploration robots, environment monitoring: the Baltic
watch project
Robot Planning TCU 16
TCU activities
ECAI Workshop August 2000
Perception Planning• Talks
– Perception planning for surveillance, tracking or cooperationP.Fabiani
– Perception planning for autonomous service robotsM.Beetz, J.Schumacher
– The WITAS project, anchoring and perception planningP.Doherty
• Discussion on representations and reasoning required for sensor models, environments, sensing tasks
Robot Planning TCU 17
TCU Activities
Dagstuhl seminar, October 2001
Plan-based Control of Robotics Agents• Topics:
– perception in plan-based control
– plan notations, plan execution, and monitoring
– execution-time plan management
– formal models of plan-based control
– plan-based control and learning
– challenge problems and benchmarks
• Participants: 50, from Europe, USA, Australia• Planned outcome: Book LNAI, Springer
Robot Planning TCU 18
TCU Activities
Workshops during 2002
• At AIPS'02– 3rd Cognitive Robotics Workshop– Other workshops ? (submission deadline: September 21st)
• At ECAI'02