I2F: An Adaptive Iterative and Inheritance Framework for Optimization-Based Trajectory Planning

This paper proposes a trajectory planner for autonomous vehicles on curvy roads. At present, the more prevailing and emerging methods are to convert the trajectory planning task into an optimal control problem (OCP) which requires satisfying two key constraints: (1) Vehicle kinematic constraints; (2) Collision avoidance constraint. Nevertheless, the dimensionality of the nominal OCP tends to be high, primarily due to the intricately formulated collision-avoidance constraints. An idea to address this issue is to first plan a coarse trajectory which guide a homotopic route, and then construct the within-corridor constraints along the coarse trajectory instead of the redundant collision-avoidance constraints. But the new challenge is how to attenuate the negative influence of the change of collision-avoidance constraints form on optimality and feasibility of the OCP. To address this issue, an iterative and inheritance framework is proposed, termed I²F. First, an intermediate OCP is formulated once per iteration, which only contains the within-corridor constraints. Second, the construction process of the within-corridor is accelerated by adopting the inheritance strategy. Through this framework, the convergence rate can be improved while reducing the cost, and weakening sensitivity of OCP to the initial guess. Finally, the efficiency and robustness of the proposed planner, along with several widely-used optimization-based planners, are thoroughly evaluated through 100 simulation cases, considering various metrics such as cost, success rate, and computational time.