package aima.core.search.online;
import java.util.HashMap;
import java.util.Set;
import aima.core.agent.Action;
import aima.core.agent.Percept;
import aima.core.agent.impl.AbstractAgent;
import aima.core.agent.impl.NoOpAction;
import aima.core.search.framework.PerceptToStateFunction;
import aima.core.search.framework.evalfunc.HeuristicFunction;
import aima.core.util.datastructure.TwoKeyHashMap;
/**
* Artificial Intelligence A Modern Approach (3rd Edition): Figure 4.24, page
* 152.<br>
* <br>
*
* <pre>
* function LRTA*-AGENT(s') returns an action
* inputs: s', a percept that identifies the current state
* persistent: result, a table, indexed by state and action, initially empty
* H, a table of cost estimates indexed by state, initially empty
* s, a, the previous state and action, initially null
*
* if GOAL-TEST(s') then return stop
* if s' is a new state (not in H) then H[s'] <- h(s')
* if s is not null
* result[s, a] <- s'
* H[s] <- min LRTA*-COST(s, b, result[s, b], H)
* b (element of) ACTIONS(s)
* a <- an action b in ACTIONS(s') that minimizes LRTA*-COST(s', b, result[s', b], H)
* s <- s'
* return a
*
* function LRTA*-COST(s, a, s', H) returns a cost estimate
* if s' is undefined then return h(s)
* else return c(s, a, s') + H[s']
* </pre>
*
* Figure 4.24 LRTA*-AGENT selects an action according to the value of
* neighboring states, which are updated as the agent moves about the state
* space.<br>
* <br>
* <b>Note:</b> This algorithm fails to exit if the goal does not exist (e.g.
* A<->B Goal=X), this could be an issue with the implementation. Comments
* welcome.
*
* @author Ciaran O'Reilly
* @author Mike Stampone
*/
public class LRTAStarAgent extends AbstractAgent {
private OnlineSearchProblem problem;
private PerceptToStateFunction ptsFunction;
private HeuristicFunction hf;
// persistent: result, a table, indexed by state and action, initially empty
private final TwoKeyHashMap<Object, Action, Object> result = new TwoKeyHashMap<Object, Action, Object>();
// H, a table of cost estimates indexed by state, initially empty
private final HashMap<Object, Double> H = new HashMap<Object, Double>();
// s, a, the previous state and action, initially null
private Object s = null;
private Action a = null;
/**
* Constructs a LRTA* agent with the specified search problem, percept to
* state function, and heuristic function.
*
* @param problem
* an online search problem for this agent to solve.
* @param ptsFunction
* a function which returns the problem state associated with a
* given Percept.
* @param hf
* heuristic function <em>h(n)</em>, which estimates the cost of
* the cheapest path from the state at node <em>n</em> to a goal
* state.
*/
public LRTAStarAgent(OnlineSearchProblem problem,
PerceptToStateFunction ptsFunction, HeuristicFunction hf) {
setProblem(problem);
setPerceptToStateFunction(ptsFunction);
setHeuristicFunction(hf);
}
/**
* Returns the search problem of this agent.
*
* @return the search problem of this agent.
*/
public OnlineSearchProblem getProblem() {
return problem;
}
/**
* Sets the search problem for this agent to solve.
*
* @param problem
* the search problem for this agent to solve.
*/
public void setProblem(OnlineSearchProblem problem) {
this.problem = problem;
init();
}
/**
* Returns the percept to state function of this agent.
*
* @return the percept to state function of this agent.
*/
public PerceptToStateFunction getPerceptToStateFunction() {
return ptsFunction;
}
/**
* Sets the percept to state function of this agent.
*
* @param ptsFunction
* a function which returns the problem state associated with a
* given Percept.
*/
public void setPerceptToStateFunction(PerceptToStateFunction ptsFunction) {
this.ptsFunction = ptsFunction;
}
/**
* Returns the heuristic function of this agent.
*/
public HeuristicFunction getHeuristicFunction() {
return hf;
}
/**
* Sets the heuristic function of this agent.
*
* @param hf
* heuristic function <em>h(n)</em>, which estimates the cost of
* the cheapest path from the state at node <em>n</em> to a goal
* state.
*/
public void setHeuristicFunction(HeuristicFunction hf) {
this.hf = hf;
}
// function LRTA*-AGENT(s') returns an action
// inputs: s', a percept that identifies the current state
@Override
public Action execute(Percept psDelta) {
Object sDelta = ptsFunction.getState(psDelta);
// if GOAL-TEST(s') then return stop
if (goalTest(sDelta)) {
a = NoOpAction.NO_OP;
} else {
// if s' is a new state (not in H) then H[s'] <- h(s')
if (!H.containsKey(sDelta)) {
H.put(sDelta, getHeuristicFunction().h(sDelta));
}
// if s is not null
if (null != s) {
// result[s, a] <- s'
result.put(s, a, sDelta);
// H[s] <- min LRTA*-COST(s, b, result[s, b], H)
// b (element of) ACTIONS(s)
double min = Double.MAX_VALUE;
for (Action b : actions(s)) {
double cost = lrtaCost(s, b, result.get(s, b));
if (cost < min) {
min = cost;
}
}
H.put(s, min);
}
// a <- an action b in ACTIONS(s') that minimizes LRTA*-COST(s', b,
// result[s', b], H)
double min = Double.MAX_VALUE;
// Just in case no actions
a = NoOpAction.NO_OP;
for (Action b : actions(sDelta)) {
double cost = lrtaCost(sDelta, b, result.get(sDelta, b));
if (cost < min) {
min = cost;
a = b;
}
}
}
// s <- s'
s = sDelta;
if (a.isNoOp()) {
// I'm either at the Goal or can't get to it,
// which in either case I'm finished so just die.
setAlive(false);
}
// return a
return a;
}
//
// PRIVATE METHODS
//
private void init() {
setAlive(true);
result.clear();
H.clear();
s = null;
a = null;
}
private boolean goalTest(Object state) {
return getProblem().isGoalState(state);
}
// function LRTA*-COST(s, a, s', H) returns a cost estimate
private double lrtaCost(Object s, Action action, Object sDelta) {
// if s' is undefined then return h(s)
if (null == sDelta) {
return getHeuristicFunction().h(s);
}
// else return c(s, a, s') + H[s']
return getProblem().getStepCostFunction().c(s, action, sDelta)
+ H.get(sDelta);
}
private Set<Action> actions(Object state) {
return problem.getActionsFunction().actions(state);
}
}