Abstract | ||
---|---|---|
We introduce a new backup operator for point-based POMDP algorithms which performs a look-ahead search at depth greater than one. We apply this operator into a new algorithm, called Stochastic Search Value Iteration (SSVI). This new algorithm relies on stochastic exploration of the environment in order to update the value function. This is in opposition with existing POMDP point-based algorithms. The underlying ideas on which SSVI is based are very similar to temporal difference learning algorithms for MDPs. In particular, SSVI takes advantage of a soft-max action selection function and of the random character of the environment itself. Empirical results on usual benchmark problems show that our algorithm performs a bit better and a bit faster than HSVI2, the state of the art algorithm. This suggests that stochastic algorithms are an alternative for solving large POMDPs. |
Year | DOI | Venue |
---|---|---|
2008 | 10.1007/978-3-540-68825-9_31 | Canadian Conference on AI |
Keywords | Field | DocType |
point-based pomdp algorithm,pomdp point-based algorithm,stochastic search value iteration,new backup operator,stochastic exploration,art algorithm,stochastic point-based algorithm,value function,stochastic algorithm,new algorithm,soft-max action selection function,value iteration,action selection,temporal difference learning,look ahead | Stochastic algorithms,Temporal difference learning,Mathematical optimization,Partially observable Markov decision process,Computer science,Algorithm,Markov decision process,Bellman equation,Operator (computer programming),Artificial intelligence,Action selection,Backup | Conference |
Volume | ISSN | ISBN |
5032.0 | 0302-9743 | 3-540-68821-8 |
Citations | PageRank | References |
0 | 0.34 | 5 |
Authors | ||
2 |
Name | Order | Citations | PageRank |
---|---|---|---|
François Laviolette | 1 | 1036 | 65.93 |
Ludovic Tobin | 2 | 57 | 4.22 |