


































Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Community
Ask the community for help and clear up your study doubts
Discover the best universities in your country according to Docsity users
Free resources
Download our free guides on studying techniques, anxiety management strategies, and thesis advice from Docsity tutors
The challenges of achieving coordination among groups of intelligent agents, focusing on cooperative agents and distributed planning. It covers planning operators, hierarchical planning, and synchronizing plans in distributed environments. The text also introduces the concept of agent societies and their application in constructing teams of agents.
Typology: Study notes
1 / 42
This page cannot be seen from the preview
Don't miss anything!
Chris BrooksDepartment of Computer ScienceUniversity of San Francisco
Department of Computer Science — University of San Francisco – p.1/
We were talking about how to achieve coordination amonggroups of intelligent agents. We’ll start by talking about groups of cooperative agents. We started with contract net, which is a nice algorithm fortask allocation.
Department of Computer Science — University of San Francisco – p.2/
A plan is a sequence of operations meant to accomplish agoal. The goal is specified declaratively: at(luggage, airport),at(brooks, airport), at(students, airport) Actions are ways of accomplishing parts of a plan They have preconditions and effects.^ Preconditions must be true to perform the action^ Effects must hold after the action is taken PutLuggageInTrunk. pre: holding(luggage) effect:in(luggage, trunk) Planning is the process of finding a sequence of actionsthat accomplishes a goal.
Department of Computer Science — University of San Francisco – p.4/
Start At ( Spare,Trunk )^ Remove(Spare,Trunk) At ( Flat,Axle )^ Remove(Flat,Axle)
At ( Spare,Ground )^ PutOn(Spare,Axle)^ At ( Flat,Axle )
At ( Spare,Axle )^ Finish
At ( Spare,Trunk ) At ( Flat,Axle )
¬ (From Russell and Norvig) Notice that planning operators can be fully described bytheir preconditions and effects. Planning involves putting these operators into a partialorder that accomplishes the conditions in the goal state. Challenges: conflicts, including disjunctive effects,allowing flexibility at run time.
Department of Computer Science — University of San Francisco – p.5/
Many times, aspects of a problem can be solvedindependently. Example: taking a trip to Peru can be decomposed into:^ Buying tickets^ Getting everyone to the airport^ Getting on the plane and flying there. I can figure out how to solve each of these problems moreor less independently. Each subproblem can be represented as an AND/ORgraph Some decisions made at runtime.^ Caveat: decisions made in one subproblem may affectpossible choices in other subproblems.
Department of Computer Science — University of San Francisco – p.7/
This is an example from amilitary logistics scenario. Resources must bemoved C1 to C2. There are many ways toaccomplish some of thesubtasks. Planners
would^
like^ to leave^ as
much^ flexibility as possible.^ Department of Computer Science — University of San Francisco – p.8/
Solution 1: Submit plans to a centralized coordinator.^ Doesn’t scale^ Agents may not be willing to share more informationthan is needed. Solution 2: broadcast top-level constraints to each other.^ This allows agents to detect whether there is atop-level conflict.^ Plans will either be totally serialized or totally parallel.
Department of Computer Science — University of San Francisco – p.10/
A better solution:^ Detect whether there is:^ No problem: all possible interactions may beinterleaved.^ No solution: plans must be serialized.^ Some solution: We then ’step down a level’ in theplans and force agents to commit to particularalternatives.^ Tradeoff: Deeper level requires more communicationand interleaving (an exponential problem), butproduces finer-grained coordination.
Department of Computer Science — University of San Francisco – p.11/
Contract net and distributed planning work for tens ofagents. How can we govern environments with thousands (ormore) agents? These are often referred to as
agent societies
Still a research area Inspiration drawn from human society, Internet-scaleprotocols.
Department of Computer Science — University of San Francisco – p.13/
Research in this area can be divided into descriptive andproscriptive domains:^ descriptive: “Given a structure or behavior on theworld, what is the outcome?”^ Proscriptive: “If a structure or behavior is enforced,what outcomes result?” There is also a vigorous debate about whetherparticipants in an Internet-level agent society should betreated as self-interested, cooperative, or a mix of the two.^ Cooperation potentially allows for more beneficialoutcomes, if participants can be trusted. Many of the same issues as P2P systems arise.
Department of Computer Science — University of San Francisco – p.14/
Avoid obstacles. If you are not holding a resource, wander randomly. If yousense ’pheromones’, weight random selection towardsthem. If you find resources, pick them up and begin droppingpheromones. Follow a beacon back home.. If you make it home, drop the resource. Over time, pheromone paths are built up between thehome and the resource.
Department of Computer Science — University of San Francisco – p.16/
Achieving macro-level behavior from microlevel rules.^ How do you guarantee outcomes? Is there an efficientway to synthesize these sorts of rules? Imposition of social norms or laws^ What outcomes can be guaranteed for a given set ofnorms or laws? What language is necessary todescribe norms or laws? Mechanisms for trust and reputation^ How can noncompliance be enforced?
Department of Computer Science — University of San Francisco – p.17/
Agents will typically have preferences over outcomes^ This is declarative knowledge about the relative valueof different states of the world.^ “I prefer ice cream to spinach” Often, the value of an outcome can be quantified (perhapsin monetary terms.) This allows the agent to compare the utility (or expectedutility) of different actions. A rational agent is one that maximizes expected utility. Self-interested agents each have their own utility function.
Department of Computer Science — University of San Francisco – p.19/
By treating participants as rational agents, we can exploittechniques from game theory and economics. Assume everyone will act to maximize their own payoff How do we structure the rules of the game so that thisbehavior leads to a desired outcome? This approach is called
mechanism design
. Department of Computer Science — University of San Francisco – p.20/