基于多重示范的智能车辆运动基元表征与序列生成

In order to solve the problem of human-like trajectory generation in different driving styles, a driving style parameter extraction-based motion primitive representation method and a corresponding motion primitive sequence generation method are proposed based on the real driving trajectory data set containing the same type of multiple demonstrations. The single motion primitive is represented by a modified dynamic motion primitive method, and the singular value decomposition is introduced to separate the main shape representation parameters and driving style parameters in the same type of trajectory set. The sequence of motion primitives is fitted with a clamped B-spline curve on the premise of correlating the parameters of the independent motion primitives. The results show that the proposed single motion primitive representation method not only guarantees the accuracy of trajectory representation, but also expands the generalized adjustment ability of primitives according to driving style. On this basis, the associated primitive sequence not only achieves the clamping of the course and position of target point, but also ensures a smooth transition between the individual primitives.