CS344M Autonomous Multiagent Systems Patrick MacAlpine Department - - PowerPoint PPT Presentation

▶
cs344m autonomous multiagent systems
SMART_READER_LITE
LIVE PREVIEW

CS344M Autonomous Multiagent Systems Patrick MacAlpine Department - - PowerPoint PPT Presentation

Jan 30, 2023 138 likes •489 views

CS344M Autonomous Multiagent Systems Patrick MacAlpine Department or Computer Science The University of Texas at Austin Good Afternoon, Colleagues Are there any questions? Patrick MacAlpine Good Afternoon, Colleagues Are there any

slide-1
SLIDE 1

CS344M Autonomous Multiagent Systems

Patrick MacAlpine Department or Computer Science The University of Texas at Austin

slide-2
SLIDE 2

Good Afternoon, Colleagues

Are there any questions?

Patrick MacAlpine

slide-3
SLIDE 3

Good Afternoon, Colleagues

Are there any questions? − Is there a deceit paradigm? − Marketing oriented approaches? − Are basis behaviors relevant? − Why does CMUnited kick ball to corners? − How does self interest benefit system as a whole?

Patrick MacAlpine

slide-4
SLIDE 4

Logistics

  • Programming assignment 3 — how was it?

Patrick MacAlpine

slide-5
SLIDE 5

Logistics

  • Programming assignment 3 — how was it?
  • Programming assignment 4

Patrick MacAlpine

slide-6
SLIDE 6

Logistics

  • Programming assignment 3 — how was it?
  • Programming assignment 4
  • Final “Exam”: Wednesday, December 9, 7:00-10:00 pm

Patrick MacAlpine

slide-7
SLIDE 7

Individual Agents

What did Sycara say about reactive vs. deliberative agents?

Patrick MacAlpine

slide-8
SLIDE 8

Individual Agents

What did Sycara say about reactive vs. deliberative agents? “Sophisticated individual agent reasoning can increase MAS coherence because each individual agent can reason about nonlocal effects of local actions, form expectations

  • f the behavior of others, or explain and possibly repair

conflicts and harmful interactions.” “Reactive agents do not have representations of their environment and act using a stimulus-response type of behavior; they respond to the present state

  • f

the environment in which they are situated.”

Patrick MacAlpine

slide-9
SLIDE 9

Individual Agents

  • Purely reactive agents have disadvantages

– Can’t react to nonlocal info or predict effects on global behavior – hard to engineer

  • Purely reactive agents have advantages

– no need to revise world model – robustness and fault tolerance if one agent fails

  • Hybrid approach better
  • Hard to evaluate agent architecture against one another

Patrick MacAlpine

slide-10
SLIDE 10

Conflicts, Resources

  • Omniscience for one agent creates bottleneck

Patrick MacAlpine

slide-11
SLIDE 11

Conflicts, Resources

  • Omniscience for one agent creates bottleneck
  • Self-interested agents: each agent maximizes own local

utility – Will that be good for global performance?

Patrick MacAlpine

slide-12
SLIDE 12

Conflicts, Resources

  • Omniscience for one agent creates bottleneck
  • Self-interested agents: each agent maximizes own local

utility – Will that be good for global performance?(invisible hand)

Patrick MacAlpine

slide-13
SLIDE 13

Conflicts, Resources

  • Omniscience for one agent creates bottleneck
  • Self-interested agents: each agent maximizes own local

utility – Will that be good for global performance?(invisible hand) – Pitfall:

Patrick MacAlpine

slide-14
SLIDE 14

Conflicts, Resources

  • Omniscience for one agent creates bottleneck
  • Self-interested agents: each agent maximizes own local

utility – Will that be good for global performance?(invisible hand) – Pitfall:tragedy of the commons – Pitfall: no stability – Pitfall: lying

Patrick MacAlpine

slide-15
SLIDE 15

Conflicts, Resources

  • Omniscience for one agent creates bottleneck
  • Self-interested agents: each agent maximizes own local

utility – Will that be good for global performance?(invisible hand) – Pitfall:tragedy of the commons – Pitfall: no stability – Pitfall: lying

  • Market-based methods/auctions

Patrick MacAlpine

slide-16
SLIDE 16

Conflicts, Resources

  • Omniscience for one agent creates bottleneck
  • Self-interested agents: each agent maximizes own local

utility – Will that be good for global performance?(invisible hand) – Pitfall:tragedy of the commons – Pitfall: no stability – Pitfall: lying

  • Market-based methods/auctions
  • Negotiation, game theory

Patrick MacAlpine

slide-17
SLIDE 17

Multiagent Planning

  • Complex individual agents
  • Teamwork modeling

– Modeling of teammates and opponents – Ad-hoc teamwork

  • Recent: emphasis on flexibility in dynamic environments

Patrick MacAlpine

slide-18
SLIDE 18

Multiagent Planning

  • Complex individual agents
  • Teamwork modeling

– Modeling of teammates and opponents – Ad-hoc teamwork

  • Recent: emphasis on flexibility in dynamic environments
  • (pursuit slides)

Patrick MacAlpine

slide-19
SLIDE 19

Pursuit Activity

Group 1: homogeneous, non-communicating Group 2: homogeneous, communicating Group 3: heterogeneous, non-communicating Group 4: heterogeneous, communicating

Patrick MacAlpine

slide-20
SLIDE 20

Communication

  • Middle agents (brokers)
  • Standard languages
  • Ontologies

More next week

Patrick MacAlpine

slide-21
SLIDE 21

Mataric: Adaptive Group Behavior

  • Built using subsumption architecture

Patrick MacAlpine

slide-22
SLIDE 22

Mataric: Adaptive Group Behavior

  • Built using subsumption architecture
  • More complex behaviors than in Brooks’ article

– Multiagent

Patrick MacAlpine

slide-23
SLIDE 23

Mataric: Adaptive Group Behavior

  • Built using subsumption architecture
  • More complex behaviors than in Brooks’ article

– Multiagent

  • Hit a complexity limit?

− (Subsumption or 3T more prevalent?)

Patrick MacAlpine

slide-24
SLIDE 24

Basis Behaviors

  • Necessary and sufficient, not “optimal”

Patrick MacAlpine

slide-25
SLIDE 25

Basis Behaviors

  • Necessary and sufficient, not “optimal”

– Task dependent – Combinations: complementary, contradictory

Patrick MacAlpine

slide-26
SLIDE 26

Basis Behaviors

  • Necessary and sufficient, not “optimal”

– Task dependent – Combinations: complementary, contradictory

  • Example: locomotion

Patrick MacAlpine

slide-27
SLIDE 27

Basis Behaviors

  • Necessary and sufficient, not “optimal”

– Task dependent – Combinations: complementary, contradictory

  • Example: locomotion

– Safe-wandering, following, dispersion, aggregation, homing

Patrick MacAlpine

slide-28
SLIDE 28

Basis Behaviors

  • Necessary and sufficient, not “optimal”

– Task dependent – Combinations: complementary, contradictory

  • Example: locomotion

– Safe-wandering, following, dispersion, aggregation, homing – What 2 compound multiagent behaviors does she describe?

Patrick MacAlpine

slide-29
SLIDE 29

Basis Behaviors

  • Necessary and sufficient, not “optimal”

– Task dependent – Combinations: complementary, contradictory

  • Example: locomotion

– Safe-wandering, following, dispersion, aggregation, homing – What 2 compound multiagent behaviors does she describe? ∗ flocking ∗ foraging

Patrick MacAlpine

slide-30
SLIDE 30

Basis Behaviors

  • Necessary and sufficient, not “optimal”

– Task dependent – Combinations: complementary, contradictory

  • Example: locomotion

– Safe-wandering, following, dispersion, aggregation, homing – What 2 compound multiagent behaviors does she describe? ∗ flocking ∗ foraging – Anything special about this domain? Or could it apply just as well to others?

Patrick MacAlpine

slide-31
SLIDE 31

Discussion

Basis behaviors for other tasks

Patrick MacAlpine

slide-32
SLIDE 32

Discussion

Basis behaviors for other tasks

  • Can human behavior be thought of as arising from a set
  • f basis behaviors?
  • What kinds of basis behaviors would they be?

Patrick MacAlpine

slide-33
SLIDE 33

Discussion

Basis behaviors for other tasks

  • Can human behavior be thought of as arising from a set
  • f basis behaviors?
  • What kinds of basis behaviors would they be?
  • Would they be the same as the ones Mataric listed?
  • Are there others?

Patrick MacAlpine

slide-34
SLIDE 34

CMUnited-98

  • The details of a complete agent
  • Details of complete UT Austin Villa 3D agent optional

reading

  • Any comments?

Patrick MacAlpine