View
2
Download
0
Category
Preview:
Citation preview
Course Objectives
• Know what it takes to make a robust autonomous multi-robot team work:– Sense/Think/Act
• Understand the important, approaches, research issues and challenges in autonomous robotics.
• Know how to program an autonomous robot team.
Rationale of the course
• Well-balanced: multi-agent theory and algorithms, experimental multi-robot examples
• Understand quantitatively natural collective phenomena
• Understand how to model, design, control, evaluate, and optimize distributed intelligent systems
• Learning to present and review a research work and digging out literature
What is this Course about
• Distributed natural and artificial systems
• Coordination algorithms
• Distributed sensing and action
• Models, simulation tools targeted to distributedintelligent systems
• Multi-robot systems
Course Overview (1 – multi-agentmethodologies)
• Introduction to Multi Agent systems and Distributed AI: Agent Architectures (logic, reactive, layered, BDI)
• Distributed rational decision making: Game theory, Voting, Auctions, Coalitions Formation.
• Agent Interaction Protocols: Coordination, Cooperation, Contract Net, Negotiation.
• Agent Communication: Coordination, Speech Acts.• Distributed Problem solving and planning: Task
Allocation, Task sharing, Result Sharing, Distributed Planning.
• Formal Methods in DAI: Modal and Temporal Logic, Knowledge and Beliefs.
Next week paper
Designing and Understanding Adaptive Group Behavior
Maja J. Matarić
Adaptive Behavior
September 1995 vol. 4 no. 1 51-80
INTRODUCTION TO MULTI-ROBOT
What we (should) got so far…
Robot:• In a real environemt
• Able to perceive and act
• Able to think
Environment
Robot
Why Multi-robot?
Why Multi-robot?
• Some tasks require a team
Why Multi - robot?
• Some tasks require a team
Why Multi-robot?
• Faster execution
• Some tasks may be decomposed ad divided in order to be executed more efficiently– Mapping of a big area
Why Multi-robot?
• Faster execution
• Some tasks may be decomposed ad divided in order to be executed more efficiently– Mapping of a big area
Why Multi-robot?
• Costs
• More specialist robots are preferred to a single generalist one– Simpler design
Why Multi-robot?
• Failure robustness is increased by the number of robots and the replication
Why Not Multi-robot?
• Coordination
• Communication
• Test difficulties
• N x Problem
Coordination
• Coordination problems are present both in artificial and natural systems.
• Examples from nature:
Communication
• Cooperation between robots can be achieved by communication mechanisms and the exchange of messages.
• Direct and indirect communication.
Communication
• Direct communication makes use of dedicated hardware
• Indirect communication makes use of stigmergy– Can reduce complexity in the
design of large scale systems
Cooperation taxonomy
Cooperation
Cooperation
• The ability of a group to cooperate in order to achieve a common goal.
• We can distinguish among cooperative and competitive systems.
• A pure cooperative system has a single shared goal among the agents.
Cooperation taxonomy
AwareUnaware
Cooperation
Knowledge
• Represents the ability of a the robots to have information about the rest of the group.
• Aware: robots are aware of the team metes• Unaware: robots acts without considering their
team mates– Frequently inspired by biological domains– Easier to manage from the engineering point of view
• Knowledge is not equal to communication: robot can manage the presence of other robots without the necessity to communicate with them.
Cooperation taxonomy
AwareUnaware
Cooperation
Strong coordination
Weakcoordination
NoCoordination
Coordination levels
• We talk about coordination when, before acting, the robot takes into account other robots actions in order to have a coherent global behavior.
• There are different methods to take into account other robots actions. – An interaction protocol is defined as a set of rules that robots have to
follows in order to interact. – May require a subdivision in different roles.– According to the protocol we can classify different coordination
mechanisms.
Coordination
What kind of coordination?
What kind of coordination?
Cooperation taxonomy
Centralizedcontrol
Distributedcontrol
Weakcentralized
control
AwareUnaware
Cooperation
Strong coordination
Weakcoordination
NoCoordination
Organization and Control
• How, in a group of robots, decisions take place?
• Main distinction among centralized and distributed control.
Centralized
• A single system takes decision for the group
Potentially optimal
Coordination may be implicit
– Difficult management
– Single point of failure
– Slow reaction time.
Centralized construction;
Khatib et al 1996
Distributed
• Each robots takes decision according to its knowledge
Simple and quick
Multiple simultaneous tasks
- Explicit coordination
CMPack 2002
Coordination
Weak Coordination Strong Coordination
Sub-tasks decomposition
Parallel execution
Minor interaction
Strategies to decompose and allocate tasks are needed.
Not decomposable tasks
Coordinated execution
Strong interaction
Tasks for multi-robot teams
• Mapping and exploration
• Target tracking
• Inspection
Weak coordination
Tasks for multi-robot teams
• Object transportation
• Robot soccer
• Large-scale construction
• Coordinated exploration
RoboticConstruction.
Strong coordination
Box Carrying
Trasporto di oggetti
Working with objects
Flocking
Human-robot Coordination
Human-robot Coordination
TAXONOMIES OF MULTI-ROBOT SYSTEMS
Importance
• comparison
• evaluate tradeoffs
• defines issues
• aids in generalizing
Sample Taxonomies
• Decker– agent granularity– heterogeneity of agent knowledge– control distribution– communication methods
• Cao et al.– group architecture– resource conflicts– origins of cooperation– learning– geometric problems
[Decker, 1995] Environment centered analysis and design of coordinationmechanisms. PhD Thesis, University of Massachusetts.[Cao et al., 1997] Cao, Y., Fukuna, A., and Kahng, A. (1997). Cooperative mobile
robotics : Antecedents and directions. Autonomous Robots, 4 :7–27.
Dudek, Jenkin & Milios
• communication– range– topology– bandwidth
• size• composition• reconfigurability• processing ability
[Dudek, 1997] Dudek, G. (1997). A taxonomy for multiagent robotics. Autonomous Robots, 3 :375–397.
Dudek, Jenkin & Milios
• range
– none
– near
– infinite
• bandwidth
– infinite
– low
– zero
• topology
– broadcast
– address
– tree
– graph
Dudek, Jenkin & Milios
• size
– alone
– pair
– limited
– infinite
• composition
– identical
– homogeneous
– heterogeneous
Dudek, Jenkin & Milios
• reconfigurability
– static
– coordinated
– dynamic
• processing ability
– summation unit
– finite state automata
– push-down automata
– turing machine
Exploration using a Topological MapDudek et al.
The collective operates by having individual robots start at a common location and then move independently to explore parts of the graph. Each robot has a unique marker which the robot can pick up/put down at its current location. The individual members meet on a pre-arranged schedule to merge their maps and subdivide the remaining territory.
Exploration using a Topological MapDudek et al.
• communication
– range
– topology
– bandwidth
• size
• composition
• reconfigurability
• processing ability
Exploration using a Topological MapDudek et al.
• communication
– range
– topology
– bandwidth
• size
• composition
• reconfigurability
• processing ability
= near
= address
= infinite
= limited
= homogeneous
= cooperative
= turing machine
Moving in FormationDudek et al.
The collective operates in a leader-follower manner in which the leader robot signals its intention to the follower robot. The signaling is performed by the leader robot making specific motions prior to the intended motion which can be easily sensed by the followers.
Moving in FormationDudek et al.
• communication
– range
– topology
– bandwidth
• size
• composition
• reconfigurability
• processing ability
Moving in FormationDudek et al.
• communication
– range
– topology
– bandwidth
• size
• composition
• reconfigurability
• processing ability
= near
= broadcast
= low
= limited
= heterogeneous
= cooperative
= turing machine
Stone & Veloso
• degree of heterogeneity
– homogeneous
– heterogeneous
• degree of communication
– non-communicating
– communicating
Stone, P., Veloso, M.M.: Multiagent systems: A survey from a machine learning perspective. Autonomous Robots 8(3), 345–383 (2000)
Predator/Prey Domain
environment
goal
Homogeneous, Non-Communicating
• reactive vs. deliberative
• local vs global perspective
• modeling other agents’ states
– recursive modeling method
• affecting others
Heterogeneous, Non-Communicating
• benevolence vs. competitiveness
– third most important issue
• fixed vs learning agents
– arms race
• modeling other agents
• roles
Homogeneous, Communicating
• distributed sensing
• communication content
• bandwidth
• topology
• range
Heterogeneous, Communicating
• understanding each other
• planning communication
– cost & freedom
• negotiation
– auctions
• commitment/decommitment
• collaborative localization
Recommended