Automobile automatic driving motion planning method and system based on learning sampling mode

Abstract

The invention relates to an automobile automatic driving motion planning method and system based on a learning sampling mode. The method comprises the steps: building an automobile kinematics model; initializing an Open table and a Closed table; calculating an evaluation value of each forward simulation trajectory, and selecting the trajectory with the highest evaluation value as a regular optimaltrajectory; performing Q value function estimation on the forward simulation trajectory, and selecting the trajectory with the maximum Q value as a reinforcement learning trajectory; selecting an initial optimal trajectory from the regular optimal trajectory and the reinforcement learning trajectory, and storing the initial optimal trajectory into a Closed table; screening non-collision forward simulation trajectories by using a collision detection method, and storing the non-collision forward simulation trajectories into an Open table; calculating an evaluation value of each forward simulation trajectory, selecting the forward simulation trajectory with the highest evaluation value as a candidate optimal trajectory, and storing the candidate optimal trajectory into a Closed table; endingthe motion planning process when the end point of the candidate optimal track is in an end point range required by motion planning; connecting the initial optimal trajectory and the candidate optimaltrajectory in the Closed table to form a final planned trajectory.

Description

technical field [0001] The invention relates to the field of intelligent vehicles, in particular to a learning sampling-based method and system for vehicle automatic driving motion planning. Background technique [0002] In recent years, artificial intelligence technology has gradually begun to be commercialized in the field of intelligent transportation and vehicles, and intelligent networked vehicles have gradually entered people's field of vision. Generally speaking, the autonomous driving software system of intelligent vehicles can be divided into four modules: perception, positioning, decision-making and control. As the most important part of the decision-making module, motion planning determines the decision-making quality of intelligent vehicles. Since the control module generally only completes the work of motion / trajectory tracking, the result of motion planning has an important impact on the final driving behavior of the vehicle. [0003] Existing motion planning...

AI Technical Summary

Problems solved by technology

The method based on end-to-end learning establishes a direct mapping from sensor data to driving actions, but due to the black-box nature of the learning method, it is difficult to carry out engineering practice and optimization; methods based on optimization generally rely on lane lines or other prior road information to solve Time is often difficult to guarantee; sampling-based methods are widely used in motion planning for autonomous driving due to their fast solution speed and ability to adapt to various environmental characteristics

[0004] Sampling-based methods generally use a cost function to select the sampling trajectory or motion state. This method is essentially an optimal trajectory / motion state selection based on artificially set rules, but the artificially set cost function is difficult to adapt to complex and complex changing reality

Images