Robot planning is needed for robots to perform purposeful missions in their environments. In realistic situations, planning does not simply involve getting to a desired destination without collisions, but often requires achieving a desired goal configuration in an optimal or near-optimal fashion. Common optimality criteria include minimum distance, minimum time, and minimum energy. To plan feasible robot motions, it is necessary to take into account robot dynamics and/or power models along with motion constraints imposed by the actuators and the environment, which are kinodynamic motion planning problems. In addition, due to the dynamic nature of the robot environments and the uncertainties in locomotion, robots must be able to plan and replan trajectories in a computationally efficient manner. This seminar reviews Sampling Based Model Predictive Optimization (SBMPO), a motion planning framework capable of addressing the needs expressed above. The seminar will present the usage of heuristics and also discuss early results on learning models and heuristics.