A method and system of a sparse sampling planner uses a finite number of measurements to determine a track's expected intermediate kinematic and classification state for a specific sensor action. It uses the expected track state to compute a reward function. The expected states are further propagated for actions at the next time step to determine the next states and so on. The sampling becomes sparse and the reward function is discounted as one propagates further in time. This produces a state-action tree that is more top-heavy while providing greater accuracy at times closer to the decision point. By doing so, the planner creates a plan comprising a sequence of actions that result in the highest reward. By employing various heuristics to further prune the tree gives highly accurate results with significant savings in computational processor time.