Methodology of hierarchical collision avoidance for high-speed self-driving vehicle based on motion-decoupled extraction of scenarios

摘要

Collision avoidance is an important requirement for self-driving systems, particularly in high-speed scenarios, where a multi-state coupled motion makes it difficult to simultaneously reach the required accuracy, efficiency, and universal feasibility for different obstacle-avoidance behaviour. For a coupled multi-state complexity, a hierarchical collision-avoidance strategy is proposed that refines the requirements for travelling under such a scenario into two levels, general and special. At the general level, the moving elliptical contour of the subject vehicle is regularised as a settled circle through a projective transformation, which attempts to determine the subject-motion-decoupled scenario. Throughout the transformation, all positional relationships between the subject and the object vehicles are retained using invariants. At the special level, a group of relative critical collision trajectories is achieved through a feature-distance-based multi-dimensional geometric optimisation model. Under the motion-decoupled scenario, a precise collision avoidance condition is constructed by mathematically expressing the relative critical collision trajectory group using a parameterised spatio-temporal curvilinear interpolation model, which provides a reasonable safety redundancy and trajectory domain to ensure both the efficiency and accuracy of the computation. In a simulation, planning trajectories using this collision-avoidance strategy is adaptive for different collision-avoidance behaviour and are more efficient than those of other algorithms.