A Survey of Trajectory Planning Methods for Autonomous Driving—Part I: Unstructured Scenarios
Yuqing Guo, Zelin Guo, Yazhou Wang, Danya Yao, Bai Li, Li Li
Abstract
Trajectory planning is a critical function in an autonomous vehicle, which is about generating a local spatio-temporal curve with safety, traverse efficiency, and comfort factors considered. Many survey papers have been published about trajectory planning in structured scenarios while a survey regarding the planners in unstructured scenarios is still absent. Driving in unstructured scenarios involves massive irregularly placed obstacles and mixed forward/backward maneuvers, which easily make the existing structured-scenario planners inapplicable. This paper aims to leverage the challenges in unstructured scenarios, review the existing planners about their strengths, limitations, and computational complexities, develop an open-source library containing prevalent trajectory planners suitable for unstructured scenarios, and provide insights into future developments. Particularly, we advocate the usage of a two-stage computational architecture, where stage one finds a coarse trajectory/path and stage two refines via a gradient-based optimizer. The two-stage architecture has been well adopted in the past score and deserves to be further developed for complex real-world trajectory planning cases in unstructured driving scenarios.