This work seeks to generate high-quality paths for complex dynamic systems. Some of the challenges addressed by this project include extending asymptotically near-optimal motion planning to systems with complex dynamics and learning better metrics for such systems. In general, providing asymptotically optimal motion planning solutions requires a boundary value problem (BVP) solver; however, this work aims to show that convergence to near-optimal solutions can be achieved without the BVP solver. This is especially important for physics-based simulations, where the system dynamics are unknown and BVP solvers for the system do not exist.