This work develops a class of stochastic optimization algorithms. It aims t
o provide numerical procedures for solving threshold-type optimal control p
roblems. The main motivation stems from applications involving optimal or s
uboptimal hedging policies, for example, production planning of manufacturi
ng systems including random demand and stochastic machine capacity. The pro
posed algorithm is a constrained stochastic approximation procedure that us
es random-direction finite-difference gradient estimates. Under fairly gene
ral conditions, the convergence of the algorithm is established and the rat
e of convergence is also derived. A numerical example is reported to demons
trate the performance of the algorithm.