Combining on- And offline optimization techniques for efficient autonomous vehicle's trajectory planning

This paper presents a general framework for autonomous trajectory planning of vehicles with a broad range of dynamic capabilities, operating in complex 3D environments, and featuring diverse mission objectives. The approach combines online receding horizon trajectory optimization and offline dynamic programming to leverage both the computational and memory resources. The online planner accounts for the immediate environment and vehicle dynamics to compute the trajectory, and the latter captures the global environment and mission features in the form of a value function that serves as terminal cost for the online trajectory optimization. The paper presents simulation results to illustrate the framework's key capabilities and demonstrate its computational tractability.