Motion Primitives Representation, Extraction and Connection for Automated Vehicle Motion Planning Applications

摘要: Developing an autonomous driving system which can generate human-like actions requires the ability to utilize basic skills learned from data. The efficiency of algorithm be significantly improved if we decompose complex tasks into motion primitives (MPs) represent elementary composition skills. Therefore, purpose this paper is MPs, extract MPs unlabeled data, and then connect in established library. By applying a probabilistic inference based on Expectation-Maximization (EM) initial segmentation, extraction method segments observed trajectories while learning set represented by modified dynamic movement (DMPs). Moreover, proposed connection transforms problem re-representation MP sequence. This demonstrates that DMP not only driver’s trajectory with acceptable accuracy but also have strong generalization ability. We present how mutual dependency between representation achieve segmentation library establishment. Besides, shows correlates independent sequence ensure smooth transition evaluates tracking accuracy. results show realizes re-generation making use interdependence relationship often neglected single MP, different types combination multiple MPs.