burlap.behavior.singleagent.planning.deterministic

Class DeterministicPlanner

java.lang.Object

burlap.behavior.singleagent.MDPSolver

burlap.behavior.singleagent.planning.deterministic.DeterministicPlanner

All Implemented Interfaces:

MDPSolverInterface, Planner

Direct Known Subclasses:

BestFirst, BFS, DFS, IDAStar

public abstract class DeterministicPlanner
extends MDPSolver
implements Planner

This class extends the OOMDPlanner to provide the interface and common mechanisms for classic deterministic forward search planners. Since classic forward search planners search for a goal state, this class makes use of a StateConditionTest object to indicate goal states. Although these planners do not strictly compute policies, but instead action sequences to be executed from some initial state, it also stores an internal partial policy that specifies which action to take in each state that is on the path of a plan it has previously found. If the same valueFunction is used multiple times from different initial states (but keep the same goal condition) it will progressively fill out the policy. If the valueFunction fails to find a valid plan, it will throw a DeterministicPlanner.PlanningFailedException runtime exception.

Author:

James MacGlashan

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	DeterministicPlanner.PlanningFailedException Exception class for indicating that a solution failed to be found by the planning algorithm.

Field Summary

Fields

Modifier and Type	Field and Description
protected StateConditionTest	gc This State condition test should return true for goal states and false for non-goal states.
protected java.util.Map<HashableState,GroundedAction>	internalPolicy Stores the action plan found by the valueFunction as a deterministic policy

Fields inherited from class burlap.behavior.singleagent.MDPSolver

actions, debugCode, domain, gamma, hashingFactory, mapToStateIndex, rf, tf

Constructor Summary

Constructors

Constructor and Description
DeterministicPlanner()

Method Summary

Methods

Modifier and Type	Method and Description
void	deterministicPlannerInit(Domain domain, RewardFunction rf, TerminalFunction tf, StateConditionTest gc, HashableStateFactory hashingFactory) Initializes the valueFunction.
protected void	encodePlanIntoPolicy(SearchNode lastVisitedNode) Encodes a solution path found by the valueFunction into this class's internal policy structure.
boolean	hasCachedPlanForState(State s) Returns whether the valueFunction has a plan solution from the provided state.
protected boolean	planContainsOption(SearchNode lastVisitedNode) Returns true if a solution path uses an option in its solution.
protected boolean	planHasDupilicateStates(SearchNode lastVisitedNode) Returns true if a solution path visits the same state multiple times.
GroundedAction	querySelectedActionForState(State s) Returns the action suggested by the valueFunction for the given state.
void	resetSolver() This method resets all solver results so that a solver can be restarted fresh as if had never solved the MDP.

Methods inherited from class burlap.behavior.singleagent.MDPSolver

addNonDomainReferencedAction, getActions, getAllGroundedActions, getDebugCode, getDomain, getGamma, getHashingFactory, getRf, getRF, getTf, getTF, setActions, setDebugCode, setDomain, setGamma, setHashingFactory, setRf, setTf, solverInit, stateHash, toggleDebugPrinting, translateAction

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface burlap.behavior.singleagent.planning.Planner

planFromState

Methods inherited from interface burlap.behavior.singleagent.MDPSolverInterface

addNonDomainReferencedAction, getActions, getDebugCode, getDomain, getGamma, getHashingFactory, getRf, getRF, getTf, getTF, setActions, setDebugCode, setDomain, setGamma, setHashingFactory, setRf, setTf, solverInit, toggleDebugPrinting

Field Detail

gc

protected StateConditionTest gc

This State condition test should return true for goal states and false for non-goal states.

internalPolicy

protected java.util.Map<HashableState,GroundedAction> internalPolicy

Stores the action plan found by the valueFunction as a deterministic policy

Constructor Detail

DeterministicPlanner

public DeterministicPlanner()

Method Detail

deterministicPlannerInit

public void deterministicPlannerInit(Domain domain,
                            RewardFunction rf,
                            TerminalFunction tf,
                            StateConditionTest gc,
                            HashableStateFactory hashingFactory)

Initializes the valueFunction. For some planners the reward function is not necessary, but for others that account for cost, like UFC, it should be provided. This method will automatically call the OOMDPPlanner init with the discount factor set to 1.

Parameters:

domain - the domain in which to plan.

rf - the reward function to use (probably should be negative to be compatible with most forward search planners)

tf - the terminal function

gc - test for goal conditions that should return true for goal states and false for non-goal states.

hashingFactory - the hashing factory to use for states.

resetSolver

public void resetSolver()

Description copied from interface: MDPSolverInterface

This method resets all solver results so that a solver can be restarted fresh as if had never solved the MDP.

Specified by:

resetSolver in interface MDPSolverInterface

Specified by:

resetSolver in class MDPSolver

hasCachedPlanForState

public boolean hasCachedPlanForState(State s)

Returns whether the valueFunction has a plan solution from the provided state.

Parameters:

s - the state to test whether a plan solution currently exists.

Returns:

true if a plan solution from the given state exists; false otherwise.

querySelectedActionForState

public GroundedAction querySelectedActionForState(State s)

Returns the action suggested by the valueFunction for the given state. If a plan including this state has not already been computed, the valueFunction will be called from this state to find one.

Parameters:

s - the state for which the suggested action is to be returned.

Returns:

The suggested action for the given state.

encodePlanIntoPolicy

protected void encodePlanIntoPolicy(SearchNode lastVisitedNode)

Encodes a solution path found by the valueFunction into this class's internal policy structure. If a state was visited more than once in the solution path, then the action used for the last occurrence of the state is used. If a null search node is passed to this method, it indicates that a plan was not found and a DeterministicPlanner.PlanningFailedException runtime exception is thrown.

Parameters:

lastVisitedNode - the last search node in the solution path, which should contain the goal state.

planContainsOption

protected boolean planContainsOption(SearchNode lastVisitedNode)

Returns true if a solution path uses an option in its solution.

Parameters:

lastVisitedNode - the last search node in the solution path, which should contain the goal state.

Returns:

true if a solution path uses an option in its solution; false otherwise.

planHasDupilicateStates

protected boolean planHasDupilicateStates(SearchNode lastVisitedNode)

Returns true if a solution path visits the same state multiple times.

Parameters:

lastVisitedNode - the last search node in the solution path, which should contain the goal state.

Returns:

true if a solution path uses an option in its solution; false otherwise.