#### DMCA

## Not all agents are equal: Scaling up distributed POMDPs for agent networks (2008)

### Cached

### Download Links

- [www.ifaamas.org]
- [www.aamas-conference.org]
- DBLP

### Other Repositories/Bibliography

Venue: | In: Proceedings of the seventh international |

Citations: | 25 - 4 self |

### Citations

191 | Taming decentralized POMDPs: Towards efficient policy computation for multiagent settings.. - Nair, Tambe, et al. - 2003 |

159 | Dynamic programming for partially observable stochastic games.
- Hansen, Bernstein, et al.
- 2004
(Show Context)
Citation Context ...int policy for general distributed POMDPs is NEXP-Complete [3]. While this negative complexity result has not deterred some researchers from continuing to pursue global optimality in the general case =-=[13, 7]-=-, a more popular approach has been to focus on tradeoffs for the sake of efficiency. Two types of tradeoffs have been examined. First, researchers have examined approximate techniques that sacrifice g... |

141 | Learning to Cooperate via Policy Search.
- Peshkin, Kim, et al.
- 2000
(Show Context)
Citation Context ...hts from FANS of differing the policy expressivity across agents, and heuristics for doing so may help in scaling up these algorithms. In related work focusing on approximate policies, Peshkin et al. =-=[11]-=- and Bernstein et al. [2] are examples of techniques that search for locally optimal policies using FSMs. Interactive POMDP (IPOMDP) model by [6] is presented as an alternative to the distributed POMD... |

130 | Framework for Sequential Planning in Multi-Agent Settings
- Gmytrasiewicz, Doshi
- 2005
(Show Context)
Citation Context ...planner can thus focus its available planning time on agents that require careful planning with more detailed plans, and less on agents which do not, whereas a distributed policy generation algorithm =-=[6]-=- can benefit from the reduced communication burden across the network. In order to realize our insights we choose to represent agent policies with finite-state machines and vary their expressivity by ... |

107 | Solving transition independent decentralized Markov decision processes.
- Becker
- 2004
(Show Context)
Citation Context ...le Markov Decision Problems (Distributed POMDPs) are ideally suited to plan for such agent networks, given their ability to plan in the presence of transitional and observational uncertainty in teams =-=[9, 5, 1, 2, 13]-=-. Unfortunately, the problem of finding the optimal joint policy for general distributed POMDPs is NEXP-Complete [3]. While this negative complexity result has not deterred some researchers from conti... |

97 | Networked distributed POMDPs: A synthesis of distributed constraint optimization and POMDPs.
- Nair, Varakantham, et al.
- 2005
(Show Context)
Citation Context ...DUCTION Many current and proposed applications of networks of agents, including mobile sensor networks, autonomous underwater vehicles, involve 100s of agents acting collaboratively under uncertainty =-=[8, 10]-=-. Distributed Partially Observable Markov Decision Problems (Distributed POMDPs) are ideally suited to plan for such agent networks, given their ability to plan in the presence of transitional and obs... |

92 | Approximate solutions for partially observable stochastic games with common payoffs.
- Emery-Montemerlo, Gordon, et al.
- 2004
(Show Context)
Citation Context ...le Markov Decision Problems (Distributed POMDPs) are ideally suited to plan for such agent networks, given their ability to plan in the presence of transitional and observational uncertainty in teams =-=[9, 5, 1, 2, 13]-=-. Unfortunately, the problem of finding the optimal joint policy for general distributed POMDPs is NEXP-Complete [3]. While this negative complexity result has not deterred some researchers from conti... |

91 | MAA*: A heuristic search algorithm for solving decentralized POMDPs.
- Szer, Charpillet, et al.
- 2005
(Show Context)
Citation Context ...le Markov Decision Problems (Distributed POMDPs) are ideally suited to plan for such agent networks, given their ability to plan in the presence of transitional and observational uncertainty in teams =-=[9, 5, 1, 2, 13]-=-. Unfortunately, the problem of finding the optimal joint policy for general distributed POMDPs is NEXP-Complete [3]. While this negative complexity result has not deterred some researchers from conti... |

90 | Bounded finite state controllers.
- Poupart, Boutilier
- 2004
(Show Context)
Citation Context ...In particular, the paper provides a new algorithm called FANS (FSM-based Agent Network Search for policies) for ND-POMDPs. FANS exploits finite state machines (FSMs) for compact policy representation =-=[2, 12]-=-, and uses three key novel ideas. First, a key insight in FANS is that not all agents in a network require the same expressivity in policy representation; hence, FANS introduces several heuristics to ... |

76 | Bounded policy iteration for decentralized POMDPs
- Bernstein, Hansen, et al.
(Show Context)
Citation Context |

40 | The complexity of decentralized control of MDPs - Bernstein, Zilberstein, et al. - 2000 |

37 | Letting loose a SPIDER on a network of POMDPs: Generating quality guaranteed policies.
- Varakantham, Marecki, et al.
- 2007
(Show Context)
Citation Context ...e sensor network domain, a domain representative of an important class of problems with networks of agents working in uncertain environments. FANS provides orders of magnitude improvement over SPIDER =-=[14]-=- as well as LIDJESP [10], two competing algorithms for ND-POMDPs, without significant loss in solution quality. Furthermore, our experimental comparison of different heuristics illustrates that smarte... |

23 |
Distributed Sensor Nets: A Multiagent Perspective.
- Lesser, Ortiz, et al.
- 2001
(Show Context)
Citation Context ...DUCTION Many current and proposed applications of networks of agents, including mobile sensor networks, autonomous underwater vehicles, involve 100s of agents acting collaboratively under uncertainty =-=[8, 10]-=-. Distributed Partially Observable Markov Decision Problems (Distributed POMDPs) are ideally suited to plan for such agent networks, given their ability to plan in the presence of transitional and obs... |