## Supervisor: (2007)

### BibTeX

@MISC{Kneebone07supervisor:,

author = {Michael Kneebone and Prof John Barnden},

title = {Supervisor:},

year = {2007}

}

### OpenURL

### Abstract

### Citations

4567 | A tutorial on hidden Markov models and selected applications in speech recognition
- Rabiner
- 1989
(Show Context)
Citation Context ...ystem’s state remains hidden, but the model is only observing instead of taking actions as well. These may be useful in future research and also aid my general understanding of the topic. A tutorial (=-=Rabiner 1989-=-) recommended to me gives a detailed explanation of the subject. 5 Timetable Date Activity November 2007 - December 2007 Finish research on clustering and MDP/POMDP conversion of PRM graph. January 20... |

4144 | Artificial Intelligence: A Modern Approach. 2nd edn - Russell, Norvig - 2003 |

2857 | Dynamic Programming - Bellman - 1957 |

892 | Planning and acting in partially observable stochastic domains
- Kaelbing, Littman, et al.
- 1995
(Show Context)
Citation Context ... (Kneebone 2007, section 2) details were given on how a PRM graph with uncertain edges (where the agent does not know a priori if they are traversable) could be represented as either an MDP or POMDP (=-=Kaelbling et al. 1996-=-) model. Each modelled the agent’s belief about edge traversability as a probability distribution over possible combinations of occluded edges. The exact form of this distribution depended upon the mo... |

587 | Principles of Artificial Intelligence - Nilsson - 1980 |

321 |
On the representation and estimation of spatial uncertainty
- Smith, Cheeseman
- 1987
(Show Context)
Citation Context ... separately allowing different localisation methods to be employed. This research differs from my own by requiring a known map prior to planning and also in the type of uncertainty; it is not a SLAM (=-=Smith and Cheeseman 1986-=-) system. I am primarily interested in map uncertainty as opposed to pose and location uncertainty. An interesting paper relating to POMDP solving (Jaulmes et al. 2007) is less related to the central ... |

241 | Learning policies for partially observable environments: Scaling up
- Littman, Cassandra, et al.
- 1995
(Show Context)
Citation Context ...tically compute the values after that. This can also be shown to produce sub-optimal policies even on small graphs and is in fact very close to how several early online POMDP algorithms such as QMDP (=-=Littman et al. 1995-=-) work. I don’t think this approach is worthwhile pursuing at this late stage in my PhD because the results so far have not indicated that it will lead to the development of a new algorithm or that it... |

165 | Incremental Pruning: A simple, fast, exact method for partially observable Markov decision processes
- Cassandra, Littman, et al.
- 1997
(Show Context)
Citation Context ... Related Work • Discuss other approaches to the problem and critique their strengths and weaknesses and analyse the similarities to my work. Discuss work on POMDP solvers such as Incremental pruning (=-=Cassandra et al. 1997-=-) and recent advances as well as motivating work for some of the PRM research such as Missiuro and Roy (2006) and other approaches which been mentioned in section 3 and previous reports. 4. Using disc... |

66 | Finding approximate pomdp solutions through belief compression - Roy, Gordon, et al. - 2005 |

64 |
Tutorial on Principal Component Analysis
- Smith
(Show Context)
Citation Context ...tions are important to several aspects of my future research. Jordan (2005) gives a very clear explanation of the Dirichlet and has well referred to. A tutorial on Principal Component Analysis (PCA) (=-=Smith 2002-=-) which is crucial in Thrun’s work on belief state compression in POMDPs (Roy et al. 2005) was also useful. My understanding of POMDP algorithms and models has also improved significantly in the past ... |

52 | The Stochastic Motion Roadmap: A Sampling Framework for Planning with Markov Motion Uncertainty - Alterovitz, Simeon, et al. - 2007 |

38 | Reinforcement learning: an introduction. Adaptive computation and machine learning - Sutton, Barto - 1998 |

38 | Bayesian Decision Problems and Markov Chains - Martin - 1967 |

33 | A prm-based motion planner for dynamically changing environments - Jaillet, Simeon - 2004 |

32 | A point-based POMDP algorithm for robot planning - Spaan, Vlassis - 2004 |

29 | Probability and Statistics, 2 nd Edition - DeGroot - 1986 |

24 | Efficient model learning for dialog management
- Doshi, Roy
- 2007
(Show Context)
Citation Context ... healthcare facility. The POMDP assumes that the users intent is hidden and “is inferred from a stream of noisy series of ambiguous observations, such as utterances from a speech recognition system” (=-=Doshi and Roy 2007-=-, section 1) This probabilistically models the uncertainty in where the user is instructing the chair to go and the POMDP is used to infer the intent. It must trade off the cost of making the wrong de... |

20 | A heuristic search algorithm that finds solutions with loops - LAO |

17 | Robot planning in partially observable continuous domains, Robotics: Science and Systems
- Porta, Spaan, et al.
- 2005
(Show Context)
Citation Context ...th probabilistic methods of solving them. One recent paper that extends the range of prob12slems that can be tackled with POMDP methods is “Robot Planning in Partially Observable Continuous Domains” (=-=Porta et al. 2005-=-). Classical POMDPs utilise a finite state space whereas many robot navigation methods, including the type I’m working on, feature continuous domains. Many approaches discretise the state space at a p... |

12 | Dirichlet processes, chinese restaurant processes and all that. Tutorial presentation at the NIPS Conference - Jordan - 2005 |

10 | A bayesian framework for optimal motion planning with uncertainty - Censi, Calisi, et al. |

5 | A formal framework for robot learning and control under model uncertainty - Jaulmes, Pineau, et al. - 2007 |

4 | Canadian traveler problem with remote sensing
- Bnaya, Felner, et al.
- 2009
(Show Context)
Citation Context ...e best cases matching the performance of the independent MDP model. 4 New developments elsewhere Another research group working on a very similar problem to our domain has recently published a paper (=-=Bnaya et al. 2009-=-) to IJCAI09. The Canadian Travelling Salesman (CTP) domain differs from the well known travelling salesman problem with the additional constraint that the state of some of the graph edges may change ... |

1 | Phd progress report iv: Extending probabilistic roadmaps for unknown kinodynamic constraints
- Kneebone
- 2007
(Show Context)
Citation Context ...f Mathematics that deals specifically with this subject. Much of my research time has been spent working on ways to solve the graph uncertainty problem that was described in the last progress report (=-=Kneebone 2007-=-). The key issues of how the agent should act when presented with a graph that it knows contains some uncertainty about some edges is still the central theme of my research. A large proportion of my r... |

1 | Doina Precup. Active learning in partially observable markov decision processes - Jaulmes, Pineau |

1 |
Phd progress report v: Extending probabilistic roadmaps for unknown kinodynamic constraints
- Kneebone
- 2007
(Show Context)
Citation Context ...iew of MDP model formulation The major theme of my research since last autumn has been in the development of the MDP formulation of PRM route graphs with uncertain edges. In the last progress report (=-=Kneebone 2007-=-, section 2) details were given on how a PRM graph with uncertain edges (where the agent does not know a priori if they are traversable) could be represented as either an MDP or POMDP (Kaelbling et al... |

1 | The perils of multi-core programming why things don’t just get faster - Kneebone |

1 | Irlab talk on incremental pruning: A simple, fast, exact method for partially observable markov decision processes, March 2008b. URL http://www.cs. bham.ac.uk/ ∼mlk/inc prune.ppt - Kneebone |

1 | Uncertain probabilistic roadmaps with observations
- Dearden, Kneebone
- 2008
(Show Context)
Citation Context ... examined in the following section. 1.1 Papers written/published As of the last progress report, a conference paper had been submitted for the PlanSIG08 conference with acceptance pending. The paper (=-=Dearden and Kneebone 2008-=-) was accepted in mid November and a half hour presentation slot at the conference was allocated. I feel this was a significant development as it is my first publication for my PhD. I gave a talk at t... |

1 |
Phd progress report vii: Extending probabilistic roadmaps for unknown obstacles
- Kneebone
- 2008
(Show Context)
Citation Context ... the following section. 1.1 Papers written/published As of the last progress report, a conference paper had been submitted for the PlanSIG08 conference with acceptance pending. The paper (Dearden and =-=Kneebone 2008-=-) was accepted in mid November and a half hour presentation slot at the conference was allocated. I feel this was a significant development as it is my first publication for my PhD. I gave a talk at t... |