Results 1  10
of
31
Solving POMDPs by Searching in Policy Space
, 1998
"... Most algorithms for solving POMDPs iteratively improve a value function that implicitly represents a policy and are said to search in value function space. This paper presents an approach to solving POMDPs that represents a policy explicitly as a finitestate controller and iteratively improve ..."
Abstract

Cited by 94 (9 self)
 Add to MetaCart
Most algorithms for solving POMDPs iteratively improve a value function that implicitly represents a policy and are said to search in value function space. This paper presents an approach to solving POMDPs that represents a policy explicitly as a finitestate controller and iteratively improves the controller by search in policy space. Two related algorithms illustrate this approach. The first is a policy iteration algorithm that can outperform value iteration in solving infinitehorizon POMDPs. It provides the foundation for a new heuristic search algorithm that promises further speedup by focusing computational effort on regions of the problem space that are reachable, or likely to be reached, from a start state. 1 Introduction A partially observable Markov decision process (POMDP) provides an elegant mathematical model for planning and control problems for which there can be uncertainty about the effects of actions and about the current state. It is wellknown that ...
Enhanced IterativeDeepening Search
, 1993
"... Iterativedeepening searches mimic a breadthfirst node expansion with a series of depthfirst searches that operate with successively extended search horizons. They have been proposed as a simple way to reduce the space complexity of bestfirst searches like A* from exponential to linear in the sea ..."
Abstract

Cited by 69 (3 self)
 Add to MetaCart
Iterativedeepening searches mimic a breadthfirst node expansion with a series of depthfirst searches that operate with successively extended search horizons. They have been proposed as a simple way to reduce the space complexity of bestfirst searches like A* from exponential to linear in the search depth. But there
Efficient memorybounded search methods
"... Memorybounded algorithms such as Korf's IDA* and Chakrabarti et al's MA* are designed to overcome the impractical memory requirements of heuristic search algorithms such as A*. It is shown that IDA * is inefficient when the heuristic function can take on a large number of values � this is a conseq ..."
Abstract

Cited by 54 (0 self)
 Add to MetaCart
Memorybounded algorithms such as Korf's IDA* and Chakrabarti et al's MA* are designed to overcome the impractical memory requirements of heuristic search algorithms such as A*. It is shown that IDA * is inefficient when the heuristic function can take on a large number of values � this is a consequence of using too little memory. Two new algorithms are developed. The first, SMA*, simpli es and improves upon MA*, making the best use of all available memory. The second, Iterative Expansion (IE), is a simple recursive algorithm that uses linear space and incurs little overhead. Experiments indicate that both algorithms perform well.
Pruning Duplicate Nodes in DepthFirst Search
 In AAAI National Conference
, 1993
"... Bestfirst search algorithms require exponential memory, while depthfirst algorithms require only linear memory. On graphs with cycles, however, depthfirst searches do not detect duplicate nodes, and hence may generate asymptotically more nodes than bestfirst searches. We present a technique for ..."
Abstract

Cited by 37 (3 self)
 Add to MetaCart
Bestfirst search algorithms require exponential memory, while depthfirst algorithms require only linear memory. On graphs with cycles, however, depthfirst searches do not detect duplicate nodes, and hence may generate asymptotically more nodes than bestfirst searches. We present a technique for reducing the asymptotic complexity of depthfirst search by eliminating the generation of duplicate nodes. The automatic discovery and application of a finite state machine (FSM) that enforces pruning rules in a depthfirst search, has significantly extended the power of search in several domains. We have implemented and tested the technique on a grid, the Fifteen Puzzle, the TwentyFour Puzzle, and two versions of Rubik's Cube. In each case, the effective branching factor of the depthfirst search is reduced, reducing the asymptotic time complexity. IntroductionThe Problem Search techniques are fundamental to artificial intelligence. Bestfirst search algorithms such as breadthfirst se...
Bidirectional Heuristic Search Reconsidered
 Journal of Artificial Intelligence Research
, 1997
"... The assessment of bidirectional heuristic search has been incorrect since it was first published more than a quarter of a century ago. For quite a long time, this search strategy did not achieve the expected results, and there was a major misunderstanding about the reasons behind it. Although there ..."
Abstract

Cited by 32 (2 self)
 Add to MetaCart
The assessment of bidirectional heuristic search has been incorrect since it was first published more than a quarter of a century ago. For quite a long time, this search strategy did not achieve the expected results, and there was a major misunderstanding about the reasons behind it. Although there is still widespread belief that bidirectional heuristic search is afflicted by the problem of search frontiers passing each other, we demonstrate that this conjecture is wrong. Based on this finding, we present both a new generic approach to bidirectional heuristic search and a new approach to dynamically improving heuristic values that is feasible in bidirectional search only. These approaches are put into perspective with both the traditional and more recently proposed approaches in order to facilitate a better overall understanding. Empirical results of experiments with our new approaches show that bidirectional heuristic search can be performed very efficiently and also with limited mem...
Anytime Heuristic Search: First Results
, 1997
"... We describe a simple technique for converting heuristic search algorithms into anytime algorithms that offer a tradeoff between search time and solution quality. The technique is related to work on use of nonadmissible evaluation functions that make it possible to find good, but possibly subop ..."
Abstract

Cited by 24 (3 self)
 Add to MetaCart
We describe a simple technique for converting heuristic search algorithms into anytime algorithms that offer a tradeoff between search time and solution quality. The technique is related to work on use of nonadmissible evaluation functions that make it possible to find good, but possibly suboptimal, solutions more quickly than it takes to find an optimal solution. Instead of
Limited discrepancy beam search
 In Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI
, 2005
"... Beam search reduces the memory consumption of bestfirst search at the cost of finding longer paths but its memory consumption can still exceed the given memory capacity quickly. We therefore develop BULB (Beam search Using Limited discrepancy Backtracking), a complete memorybounded search method th ..."
Abstract

Cited by 17 (1 self)
 Add to MetaCart
Beam search reduces the memory consumption of bestfirst search at the cost of finding longer paths but its memory consumption can still exceed the given memory capacity quickly. We therefore develop BULB (Beam search Using Limited discrepancy Backtracking), a complete memorybounded search method that is able to solve more problem instances of large search problems than beam search and does so with a reasonable runtime. At the same time, BULB tends to find shorter paths than beam search because it is able to use larger beam widths without running out of memory. We demonstrate these properties of BULB experimentally for three standard benchmark domains. 1
Stochastic Node Caching for Memorybounded Search
, 1998
"... Linearspace search algorithms such as IDA* (Iterative Deepening A*) cache only those nodes on the current search path, but may revisit the same node again and again. This causes IDA* to take an impractically long time to find a solution. In this paper, we propose a simple and effective algorit ..."
Abstract

Cited by 14 (1 self)
 Add to MetaCart
Linearspace search algorithms such as IDA* (Iterative Deepening A*) cache only those nodes on the current search path, but may revisit the same node again and again. This causes IDA* to take an impractically long time to find a solution. In this paper, we propose a simple and effective algorithm called Stochastic Node Caching (SNC) for reducing the number of revisits. SNC caches a node with the best estimate, which is currently known of the minimum estimated cost from the node to the goal node. Unlike previous related research such as MREC, SNC caches nodes selectively, based on a fixed probability. We demonstrate that SNC can effectively reduce the number of revisits compared to MREC, especially when the statespace forms a lattice. Introduction Linearspace search algorithms such as IDA* (Korf 1985) perform a series of depthfirst search iterations, gradually extending the search depth. Since they cache only nodes on the current search path, the amount of memory requir...
ITS: An Efficient LimitedMemory Heuristic Tree Search Algorithm
 In Proc. Twelfth National Conference on Artificial Intelligence (AAAI94
, 1994
"... This paper describes a new admissible tree search algorithm called Iterative Threshold Search (ITS). ITS can be viewed as a muchsimplified version of MA* [1], and a generalized version of MREC [12]. We also present the following results: 1. Every node generated by ITS is also generated by IDA*, eve ..."
Abstract

Cited by 13 (3 self)
 Add to MetaCart
This paper describes a new admissible tree search algorithm called Iterative Threshold Search (ITS). ITS can be viewed as a muchsimplified version of MA* [1], and a generalized version of MREC [12]. We also present the following results: 1. Every node generated by ITS is also generated by IDA*, even if ITS is given no more memory than IDA*. In addition, there are trees on which ITS generates O(N) nodes in comparison to O(N log N) nodes generated by IDA*, where N is the number of nodes eligible for generation by A*. 2. Experimental tests show that if the nodegeneration time is high (as in most practical problems), ITS can provide significant savings in both number of node generations and running time. Our experimental results also suggest that in the average case both IDA* and ITS are asymptotically optimal on the traveling salesman problem. Introduction Although A* is usually very efficient in terms of number of node expansions [2], it requires an exponential amount of memory, and...