1. Generalizing Plans to New Environments in Multiagent Relational MDPs Carlos Guestrin Daphne Koller Stanford University

2. Multiagent Coordination Examples Search and rescue Factory management Supply chain Firefighting Network routing Air traffic control Multiple, simultaneous decisions Exponentially-large spaces Limited observability Limited communication

3. peasant footman building Real-time Strategy Game Peasants collect resources and build Footmen attack enemies Buildings train peasants and footmen

4. Scaling up by Generalization Exploit similarities between world elements Generalize plans: From a set of worlds to a new, unseen world Avoid need to replan Tackle larger problems Formalize notion of “similar” elements Compute generalizable plans

5. Relational Models and MDPs Classes: Peasant, Gold, Wood, Barracks, Footman, Enemy… Relations Collects, Builds, Trains, Attacks… Instances Peasant1, Peasant2, Footman1, Enemy1… Value functions in class level Objects of the same class have same contribution to value function Factored MDP equivalents of PRMs [Koller, Pfeffer ‘98]

6. Relational MDPs Class-level transition probabilities depends on: Attributes; Actions; Attributes of related objects Class-level reward function Instantiation (world) Number objects; Relations Well-defined MDP Peasant P’ P APAP Gold G’ G Collects

7. Planning in a World Long-term planning by solving MDP # states exponential in number of objects # actions exponential Efficient approximation by exploiting structure! RMDP world is a factored MDP

8. Roadmap to Generalization Solve 1 world Compute generalizable value function Tackle a new world

9. World is a Factored MDP P F E G R F’ E’ G’ P’ State Dynamics Decisions Rewards P(F’|F,G,H,A F ) H APAP AFAF

10. Long-term Utility = Value of MDP Value computed by linear programming: One variable V (x) for each state One constraint for each state x and action a Number of states and actions exponential! [Manne `60]

11. Approximate Value Functions Linear combination of restricted domain functions [Bellman et al. `63] [Tsitsiklis & Van Roy `96] [Koller & Parr `99,`00] [Guestrin et al. `01] Each V o depends on state of object and related objects: State of footman Status of barracks Must find V o giving good approximate value function

12. Single LP Solution for Factored MDPs Variables for each V o, for each object Polynomially many LP variables One constraint for every state and action  Exponentially many LP constraints V o, Q o depend on small sets of variables/actions Exploit structure as in variable elimination [Guestrin, Koller, Parr `01] [Schweitzer and Seidmann ‘85]

13. Representing Exponentially Many Constraints Exponentially many linear = one nonlinear constraint

14. Can use variable elimination to maximize over state space: [Bertele & Brioschi ‘72] Variable Elimination A D BC As in Bayes nets, maximization is exponential in tree-width Here we need only 23, instead of 63 sum operations ),(),(),(max 121,, CBgCAfBAf CBA   ),(),( ),(),( 4321,, DBfDCfCAfBAf DCBA  ),(),(),(),( 4321,,, DBfDCfCAfBAf DCBA 

15. Representing the Constraints Functions are factored, use Variable Elimination to represent constraints: Number of constraints exponentially smaller

16. Roadmap to Generalization Solve 1 world Compute generalizable value function Tackle a new world

17. Generalization Sample a set of worlds Solve a linear program for these worlds: Obtain class value functions When faced with new problem: Use class value function No re-planning needed

18. Worlds and RMDPs Meta-level MDP: Meta-level LP:

19. Class-level Value Functions Approximate solution to meta-level MDP Linear approximation Value function defined in the class level All instances use same local value function

20. Class-level LP Constraints for each world represented by factored LP Number of worlds exponential or infinite Sample worlds from P(  )

21. Theorem Exponentially (infinitely) many worlds ! need exponentially many samples? NO! samples Value function within , with prob. at least 1- . R max is the maximum class reward Proof method related to [de Farias, Van Roy ‘ 02]

22. LP with sampled worlds Solve LP for sampled worlds Use Factored LP for each world Obtain class-level value function New world: instantiate value function and act

23. Learning Classes of Objects Which classes of objects have same value function? Plan for sampled worlds individually Use value function as “training data” Find objects with similar values Include features of world Used decision tree regression in experiments

24. Summary of Generalization Algorithm 1.Model domain as Relational MDPs 2.Pick local object value functions V o 3.Learn classes by solving some instances 4.Sample set of worlds 5.Factored LP computes class-level value function

25. A New World When faced with a new world , value function is: Q function becomes: At each state, choose action maximizing Q(x,a) Number of actions is exponential! Each Q C depends only on a few objects!!!

26. Q(A 1,…,A 4, X 1,…,X 4 ) ¼ Q 1 (A 1, A 4, X 1,X 4 ) + Q 2 (A 1, A 2, X 1,X 2 ) + Q 3 (A 2, A 3, X 2,X 3 ) + Q 4 (A 3, A 4, X 3,X 4 ) Local Q function Approximation M4M4 M1M1 M3M3 M2M2 Q3Q3 Q(A 1,…,A 4, X 1,…,X 4 ) Associated with Agent 3 Limited observability: agent i only observes variables in Q i Observe only X 2 and X 3 Must choose action to maximize  i Q i

27. Use variable elimination for maximization: [Bertele & Brioschi ‘72] Maximizing  i Q i : Coordination Graph Limited communication for optimal action choice Comm. bandwidth = induced width of coord. graph A1A1 A4A4 A2A2 A3A3 ),(),(),(max 321312211,, 321 AAgAAQAAQ AAA   ),(),( ),(),( 424433312211,, 4321 AAQAAQAAQAAQ AAAA  ),(),(),(),( 424433312211,,, 4321 AAQAAQAAQAAQ AAAA  If A 2 attacks and A 3 defends, then A 4 gets $10

28. Summary of Algorithm 1.Model domain as Relational MDPs 2.Factored LP computes class-level value function 3.Reuse class-level value function in new world

29. Experimental Results SysAdmin problem Unidirectional Ring Server Star Ring of Rings

30. Generalizing to New Problems

33. Classes of Objects Discovered Learned 3 classes Server Intermediate Leaf

34. Learning Classes of Objects

36. Results 2 Peasants, Gold, Wood, Barracks, 2 Footman, Enemy Reward for dead enemy About 1 million of state/action pairs Solve with Factored LP Some factors are exponential Coordination graph for action selection [with Gearhart and Kanodia]

37. Generalization 9 Peasants, Gold, Wood, Barracks, 3 Footman, Enemy Reward for dead enemy About 3 trillion of state/action pairs Instantiate generalizable value function At run-time, factors are polynomial Coordination graph for action selection

38. The 3 aspects of this talk Scaling up collaborative multiagent planning Exploiting structure Generalization Factored representation and algorithms Relational MDP, Factored LP, coordination graph Freecraft as a benchmark domain

39. Conclusions RMDP Compact representation for set of similar planning problems Solve single instance with factored MDP algorithms Tackle sets of problems with class-level value functions Efficient sampling of worlds Learn classes of value functions Generalization to new domains Avoid replanning Solve larger, more complex MDPs