burlap.oomdp.singleagent.pomdp.beliefstate

Interface BeliefState

All Superinterfaces:

State

All Known Implementing Classes:

TabularBeliefState

public interface BeliefState
extends State

An interface for defining a belief state, which is a probability distribution over MDP states. This interface does not require enumerating all states, because it is possible to have a belief state over an infinite number of MDP states. However, it does require that the probability density function be returnable (belief(burlap.oomdp.core.states.State), to be able to sample an MDP state from the belief distribution sampleStateFromBelief(), and a mechanism to update the belief state with respect to some observation and action getUpdatedBeliefState(burlap.oomdp.core.states.State, burlap.oomdp.singleagent.GroundedAction).

Author:

James MacGlashan and Nakul Gopalan

Method Summary

Methods

Modifier and Type	Method and Description
double	belief(State s) Returns the probability density/mass for the input MDP state.
BeliefState	getUpdatedBeliefState(State observation, GroundedAction ga) Computes a new belief distribution using this BeliefState as the prior and conditioned on the given POMDP observation and action taken.
State	sampleStateFromBelief() Samples an MDP state state from this belief distribution.

Methods inherited from interface burlap.oomdp.core.states.State

addAllObjects, addObject, copy, getAllObjects, getAllObjectsByClass, getAllUnsetAttributes, getCompleteStateDescription, getCompleteStateDescriptionWithUnsetAttributesAsNull, getFirstObjectOfClass, getObject, getObjectClassesPresent, getObjectMatchingTo, getObjectsOfClass, getPossibleBindingsGivenParamOrderGroups, numTotalObjects, removeAllObjects, removeObject, removeObject, renameObject, renameObject, setObjectsValue

Method Detail

belief

double belief(State s)

Returns the probability density/mass for the input MDP state.

Parameters:

s - the the input MDP state defined by a State instance.

Returns:

the probability density/mass of the input MDP state in this belief distribution.

sampleStateFromBelief

State sampleStateFromBelief()

Samples an MDP state state from this belief distribution.

Returns:

an MDP state defined by a State instance.

getUpdatedBeliefState

BeliefState getUpdatedBeliefState(State observation,
                                GroundedAction ga)

Computes a new belief distribution using this BeliefState as the prior and conditioned on the given POMDP observation and action taken.

Parameters:

observation - the conditioned POMDP observation defined by a State instance.

ga - the conditioned action selection in the previous time step.

Returns:

the new belief state distribution represented by a new BeliefState instance.