"Decision-theoretic and game-theoretic approaches to decision making in collectives", Matthijs Spaan, 11 de Março de 2009, Ciclo de Conferências "Das Sociedades Humanas às Sociedades Artificiais" (edição 200), Instituto de Sistemas e Robótica, Instituto Superior Técnico, Lisboa
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
Decision-theoretic and game-theoretic approaches to decision making in collectives (Matthijs Spaan)
1. Decision-theoretic and game-theoretic approaches
to decision making in collectives
Matthijs Spaan
Institute for Systems and Robotics
´
Instituto Superior Tecnico
Lisbon, Portugal
`
Das Sociedades Humanas as Sociedades Artificiais,
March 11, 2009
PÓLO DO I.S.T 1/26
2. Introduction
• Our research goal: “intelligent” decision making for robots
in collectives.
• Here: intelligent = rational given a task.
• Examples:
◮ Robots to assist people;
◮ Multi-robot search and rescue;
◮ Robots playing football;
◮ Software agents on the internet.
PÓLO DO I.S.T 2/26
3. Artificial Societies vs. Human Societies
• Our perspective: design artificial systems with a purpose.
• Robot societies need not resemble human societies.
• Humans: not fully rational.
• Robots: hopefully more rational.
PÓLO DO I.S.T 3/26
4. Rational robots
• Rational robots do the right thing.
◮ Knowledge of the environment.
◮ Actions the robot can take.
◮ Perceptions seen so far.
◮ Performance measure.
PÓLO DO I.S.T 4/26
7. Rational behavior
• Don’t program how the robot should behave.
• Program its environment, actions, perceptions, performance
measure.
⇒ Robot will behave rationally.
• Equip robot with tools to behave rationally.
PÓLO DO I.S.T 7/26
8. Utility
• Utility: degree of happiness.
• If a robot prefers one situation over another one, it has
higher utility.
• Utility function: utility of each situation.
• We define our robot’s utility function.
PÓLO DO I.S.T 8/26
9. Utility
Gold More Gold
Recharge
Point Lion
PÓLO DO I.S.T 9/26
20. Uncertainty
• Things don’t always work out as you want. . .
• Robots need to deal with noise.
• Uncertainty complicates maximizing utility.
PÓLO DO I.S.T 20/26
21. Maximizing Utility under Uncertainty
safe path
dangerous path
+10
+20
−20
−20
−20
PÓLO DO I.S.T 21/26
22. Robot collectives
• A collective of robots brings interesting challenges:
◮ How do they interact?
◮ Can they communicate?
◮ Are they cooperating or competing?
◮ Do all robots have identical capabilities?
• Cooperating robots ⇒ decentralized decision theory.
• Non-cooperating robots ⇒ game theory.
PÓLO DO I.S.T 22/26
25. Decentralized decision theory at ISR
• Robots have noisy, limited sensors.
• No or limited communication.
?
?
PÓLO DO I.S.T 25/26
26. Conclusions
• In contrast with humans, but we design our robots to be
rational.
• Maximizing expected utility ⇒ rational behavior.
• In reality, we can only achieve limited rationality.
• Robot societies can be designed to maximize social welfare.
• Human societies, on the other hand. . .
PÓLO DO I.S.T 26/26