Abstract
We propose a steepest descent method to compute optimal control parameters for balancing between multiple performance objectives in stateless stochastic scheduling, wherein the scheduling decision is effected by a simple constant-time coin toss operation only. We apply our method to the scheduling of a mobile sensor's coverage time among a set of points of interest (PoIs). The coverage algorithm is guided by a Markov chain, wherein the sensor at PoI i decides to go to the next PoI j with transition probability p(ij). We use steepest descent to compute the transition probabilities for optimal tradeoff among different performance goals with regard to the distributions of per-PoI coverage times and exposure times and the entropy and energy efficiency of sensor movement. For computational efficiency, we show how we can optimally adapt the step size in steepest descent to achieve fast convergence. However, we found that the structure of our problem is complex, because there may exist surprisingly many local optima in the solution space, causing basic steepest descent to easily get stuck at a local optimum. To solve the problem, we show how proper incorporation of noise in the search process can get us out of the local optima with high probability. We provide simulation results to verify the accuracy of our analysis and show that our method can converge to the globally optimal control parameters under different assigned weights to the performance goals and different initial parameters.