Autonomous unmanned vehicles require the ability to maneuver within their environments safely and efficiently. The problem of generating these traversable, collision-free paths is known as motion planning. While much work has been done in this field, the majority of the existing motion planning algorithms either fail to consider the kinodynamic constraints of the vehicle or is too computationally intensive to be performed online. The presented motion planning algorithm mitigates these issues by using kinodynamically feasible motion primitives to enforce vehicle constraints, while exploiting the computational efficiency and scalability of probabilistic roadmap planners. The result is an online motion planner that is capable of generating safe, traversable paths to any specified destination in an unknown 3D environment. This approach has been experimentally verified using quadrotor helicopters and skid-steer ground rovers.