Example #1
0
        public static CUDDNode ProbReachReward(CUDDNode trans, CUDDNode stateReward, CUDDNode transReward, CUDDNode reach, CUDDVars allRowVars, CUDDVars allColVars, CUDDNode maybe, CUDDNode inf)
        {
            CUDD.Ref(trans, maybe);
            CUDDNode a = CUDD.Function.Times(trans, maybe);

            CUDD.Ref(stateReward, maybe);
            CUDDNode newStateReward = CUDD.Function.Times(stateReward, maybe);

            CUDD.Ref(transReward, a);
            CUDDNode newTransRewards = CUDD.Function.Times(transReward, a);
            newTransRewards = CUDD.Abstract.SumAbstract(newTransRewards, allColVars);

            CUDDNode allReward = CUDD.Function.Plus(newStateReward, newTransRewards);

            //
            CUDDNode tmp = CUDD.Matrix.Identity(allRowVars, allColVars);
            CUDD.Ref(reach);
            tmp = CUDD.Function.And(tmp, reach);
            a = CUDD.Function.Minus(tmp, a);

            CUDDNode sol = Jacobi(a, allReward, allReward, reach, allRowVars, allColVars, 1);

            // set reward for infinity states to infinity
            CUDD.Ref(inf);
            sol = CUDD.Function.ITE(inf, CUDD.PlusInfinity(), sol);

            //
            CUDD.Deref(a, allReward);
            return sol;
        }
Example #2
0
        public CUDDNode ComputeReachReward(CUDDNode b, int rewardStructIndex)
        {
            //Reference to b in the calling place and create a new copy
            CUDD.Ref(b, reach);
            b = CUDD.Function.And(b, reach);

            //
            CUDDNode inf, maybe, reward;

            if (b.Equals(CUDD.ZERO))
            {
                CUDD.Ref(reach);
                inf = reach;

                maybe = CUDD.Constant(0);
            }
            else if (b.Equals(reach))
            {
                inf   = CUDD.Constant(0);
                maybe = CUDD.Constant(0);
            }
            else
            {
                CUDDNode no = ProbAlgo.Prob0(trans01, reach, allRowVars, allColVars, reach, b);

                CUDDNode prob1 = ProbAlgo.Prob1(trans01, reach, allRowVars, allColVars, reach, b, no);
                CUDD.Deref(no);

                CUDD.Ref(reach);
                inf = CUDD.Function.And(reach, CUDD.Function.Not(prob1));

                CUDD.Ref(reach, inf, b);
                maybe = CUDD.Function.And(reach, CUDD.Function.Not(CUDD.Function.Or(inf, b)));
            }

            // print out yes/no/maybe
            Debug.WriteLine("goal = " + CUDD.GetNumMinterms(b, allRowVars.GetNumVars()));
            Debug.WriteLine("inf = " + CUDD.GetNumMinterms(inf, allRowVars.GetNumVars()));
            Debug.WriteLine("maybe = " + CUDD.GetNumMinterms(maybe, allRowVars.GetNumVars()));

            if (maybe.Equals(CUDD.ZERO))
            {
                CUDD.Ref(inf);
                reward = CUDD.Function.ITE(inf, CUDD.PlusInfinity(), CUDD.Constant(0));
            }
            else
            {
                reward = ProbAlgo.ProbReachReward(trans, stateRewards[rewardStructIndex], transRewards[rewardStructIndex], reach, allRowVars, allColVars, maybe, inf);
            }

            CUDD.Deref(inf, maybe, b);

            CUDD.Ref(start);
            return(CUDD.Function.Times(reward, start));
        }
Example #3
0
        public static CUDDNode NondetInstReward(CUDDNode trans, CUDDNode stateReward, CUDDNode nondetMask, CUDDVars allRowVars, CUDDVars allColVars, CUDDVars nondetVars, int bound, bool min, CUDDNode init)
        {
            DateTime startTime = DateTime.Now;

            CUDDNode newNondetMask, sol, tmp;

            CUDD.Ref(nondetMask);
            newNondetMask = CUDD.Function.ITE(nondetMask, CUDD.PlusInfinity(), CUDD.Constant(0));

            CUDD.Ref(stateReward);
            sol = stateReward;

            for (int i = 0; i < bound; i++)
            {
                tmp = CUDD.Matrix.MatrixMultiplyVector(trans, sol, allRowVars, allColVars);

                if (min)
                {
                    CUDD.Ref(newNondetMask);
                    tmp = CUDD.Function.Maximum(tmp, newNondetMask);

                    tmp = CUDD.Abstract.MinAbstract(tmp, nondetVars);
                }
                else
                {
                    tmp = CUDD.Abstract.MaxAbstract(tmp, nondetVars);
                }

                //
                CUDD.Deref(sol);
                sol = tmp;
            }

            DateTime endTime     = DateTime.Now;
            double   runningTime = (endTime - startTime).TotalSeconds;

            Debug.WriteLine("NondetInstReward: " + bound + " iterations in " + runningTime + " seconds");

            //
            CUDD.Deref(newNondetMask);
            return(sol);
        }
Example #4
0
        /// <summary>
        /// [ REFS: '', DEREFS:]
        /// </summary>
        /// <param name="allProb"></param>
        /// <param name="state"></param>
        public double GetMinProb(CUDDNode allProb, CUDDNode filter)
        {
            double result = 0;

            CUDD.Ref(filter, reach);
            CUDDNode newFilter = CUDD.Function.And(filter, reach);

            if (newFilter.Equals(CUDD.ZERO))
            {
                result = double.MaxValue;
            }
            else
            {
                CUDD.Ref(newFilter, allProb);
                CUDDNode tmp = CUDD.Function.ITE(newFilter, allProb, CUDD.PlusInfinity());

                result = CUDD.FindMin(tmp);
                CUDD.Deref(tmp);
            }

            CUDD.Deref(newFilter);

            return(result);
        }
Example #5
0
        public static CUDDNode NondetReachReward(CUDDNode trans, CUDDNode reach, CUDDNode stateReward, CUDDNode transReward, CUDDNode nondetMask, CUDDVars allRowVars, CUDDVars allColVars, CUDDVars nondetVars, CUDDNode infReward, CUDDNode maybeReward, bool min)
        {
            DateTime startTime          = DateTime.Now;
            int      numberOfIterations = 0;

            CUDDNode a, allReward, newNondetMask, sol, tmp;

            // filter out rows (goal states and infinity states) from matrix
            CUDD.Ref(trans, maybeReward);
            a = CUDD.Function.Times(trans, maybeReward);

            // also remove goal and infinity states from state rewards vector
            CUDD.Ref(stateReward, maybeReward);
            CUDDNode tempStateReward = CUDD.Function.Times(stateReward, maybeReward);

            // multiply transition rewards by transition probs and sum rows
            // (note also filters out unwanted states at the same time)
            CUDD.Ref(transReward, a);
            CUDDNode tempTransReward = CUDD.Function.Times(transReward, a);

            tempTransReward = CUDD.Abstract.SumAbstract(tempTransReward, allColVars);

            // combine state and transition rewards
            allReward = CUDD.Function.Plus(tempStateReward, tempTransReward);

            // need to change mask because rewards are not necessarily in the range 0..1
            CUDD.Ref(nondetMask);
            newNondetMask = CUDD.Function.ITE(nondetMask, CUDD.PlusInfinity(), CUDD.Constant(0));

            // initial solution is infinity in 'inf' states, zero elsewhere
            // note: ok to do this because cudd matrix-multiply (and other ops)
            // treat 0 * inf as 0, unlike in IEEE 754 rules
            CUDD.Ref(infReward);
            sol = CUDD.Function.ITE(infReward, CUDD.PlusInfinity(), CUDD.Constant(0));

            while (true)
            {
                numberOfIterations++;

                tmp = CUDD.Matrix.MatrixMultiplyVector(a, sol, allRowVars, allColVars);

                // add rewards
                CUDD.Ref(allReward);
                tmp = CUDD.Function.Plus(tmp, allReward);

                if (min)
                {
                    CUDD.Ref(newNondetMask);
                    tmp = CUDD.Function.Maximum(tmp, newNondetMask);
                    tmp = CUDD.Abstract.MinAbstract(tmp, nondetVars);
                }
                else
                {
                    tmp = CUDD.Abstract.MaxAbstract(tmp, nondetVars);
                }

                CUDD.Ref(infReward);
                tmp = CUDD.Function.ITE(infReward, CUDD.PlusInfinity(), tmp);

                if (CUDD.IsEqual(tmp, sol))
                {
                    CUDD.Deref(tmp);
                    break;
                }

                CUDD.Deref(sol);
                sol = tmp;
            }

            DateTime endTime     = DateTime.Now;
            double   runningTime = (endTime - startTime).TotalSeconds;

            Debug.WriteLine("NondetReachReward: " + numberOfIterations + " iterations in " + runningTime + " seconds");

            //
            CUDD.Deref(a, allReward, newNondetMask);
            return(sol);
        }