information sharing for distributed planning

AAMAS 2010 - Doctoral Symposium 1

Information Sharing forDistributed Planning

Prasanna Velagapudi


Large Heterogeneous Teams

• 100s to 1000s of robots, agents, people

• Complex, collaborative tasks

• Dynamic, uncertain environment

• Joint planning intractable


Scaling Team Planning

• Independent planners: can’t account for teammates• Existing work: needs specific structure or doesn’t

scale to these sizes– DPC, Prioritized Planning– JESP, Factored MDP, ND-POMDP


Iterated Distributed Planning

1. Factor the problem, enumerate interactions2. Compute independent plans & potential interactions3. Exchange messages about interactions4. Use exchanged information, improve local model




?


A Tale of Two Distributed Planners

Distributed Prioritized Planning (DPP) L-TREMOR


Distributed Prioritized Planning


Multiagent Path Planning

Start

Goal


Multiagent Path Planning


Prioritized Planning

• Assign priorities to agents based on path length

[van den Berg, et al 2005]


Prioritized Planning

• Plan from highest priority to lowest priority• Use previous agents as dynamic obstacles

[van den Berg, et al 2005]


Distributed Prioritized Planning

Parallelizable& Equivalent


Large-Scale Path Solutions


DPP Results

Fewer Sequential Plans


DPP Results

Longer Planning TimeFewer Sequential Plans


• Prioritized Planning

• DPP

Why does this happen?

ABCD

ABCD

Longest planning agents might replan multiple times

Individual agent planning times varied by >2 orders of magnitude

Solution 2: Incremental Planning

Solution 1: Prioritize by plan time?


Summary of DPP

• Observable, certain world• Only one type of interaction: collision

• Far fewer sequential planning iterations• Incremental planning may reduce execution time


L-TREMOR


A Simple Rescue Domain

Rescue Agent

Cleaner Agent

Narrow Corridor

Victim

Unsafe Cell

Clearable Debris


A Simple (Large) Rescue Domain


Distributed POMDP with Coordination Locales (DPCL)

• Often, interactions between agents are sparse

Only fits one agent Passable if

cleaned

[Varakantham, et al 2009]



• Define coordination locales (CLs) where POMDP model functions are not independent:


<S, A, Ω, P, R, O> (states) (actions) (obs.) (transition)(reward)(obs. fn)





S1, A1 S2, A2

SglobalR1, P1, O1 R2, P2, O2

Outside CL:(typical)





S1, A1 S2, A2

Sglobal

R12, P12, O12

Inside CL:(interaction)


TREMOR

Role Allocation Policy Solution Interaction Detection Coordination

TREMOR

Branch & Bound MDP

Independent EVA[3] solvers

Joint policy evaluation

Reward shapingof independent

models



L-TREMOR

Role Allocation Policy Solution Interaction Detection Coordination

TREMOR

Branch & Bound MDP

Independent EVA[3] solvers

Joint policy evaluation Reward shaping

of independentmodels

L-TREMOR

DecentralizedAuction

Sampling & message passing

Distributed & Parallelizable


Preliminary Results – Joint Utility

N = 6 N = 10N = 100

(structurally similar to N=10)


Preliminary Results – Timing


Preliminary Results – Model Accuracy

R = 0.804


Current Issues

• Oscillations in solutions

• Discovery of relevant locales

?


Summary of L-TREMOR

• Partially-observable, uncertain world• Multiple types of interactions• Role-allocation of tasks

• Improvement over independent planning• Handles large problems• Next steps: improving convergence


Conclusions

• Two approaches to distributed planning– DPP: approaching centralized performance– L-TREMOR: exceeding joint tractability

• Analogous strategies for distributing planning– Both iterate independent planners– Both exchange messages about states, actions


Future Work

• Generalized framework for distributed planning through iterative message exchange

• Reduce necessary communication• Better search over task allocations• Scaling to larger team sizes

information sharing for distributed planning

Documents

doctoral symposium8

doctoral symposiumsolution

doctoral symposiumadd

related work

intractable aamas

local model

prioritized planningaamas

prioritized planningjesp