Rule and learning model-based highway leaving method for driverless cars

Abstract

The invention relates to a rule and learning model-based highway leaving method for driverless cars. The method comprises the steps of: in a process that a driverless car runs on a highway, generatingan off-ramp motivation at a distance before a ramp according to a navigation system, trying to get off the ramp by utilizing a rule model, judging whether off-ramp decided by a rule-based decision making model reduces the success rate or not, if the judging result is negative, adopting an action decided by the rule model, and otherwise, entering the next step; and establishing a rule and reinforcement learning mixed hybrid decision making model and a training method thereof on the basis of a framework of reinforcement learning, wherein the hybrid decision model is capable of adopting a rule model for running when being far away from the ramp, and utilizing the reinforcement learning decision making model to adjust the vehicle action according to off-ramp urgency in the process of drivingto the ramp. The method is capable of enhancing the running efficiency and stability of driverless cars in the off-ramp process, and realizing efficient and high-stability off-ramp decision making ofdriverless cars under the condition that the perception range is limited and the environment vehicles are difficult to predict.

Description

technical field [0001] The invention relates to the technical field of unmanned vehicle decision-making, in particular to a method for unmanned vehicles driving away from high speeds based on rules and learning models. Background technique [0002] The autonomous decision-making of unmanned vehicles is an important part of the unmanned vehicle system. Expressways are an important application scenario for unmanned vehicles. The driving efficiency of the car has an important impact. Changing to the rightmost lane too early to wait for the off-ramp or missing the ramp will significantly reduce the driving efficiency. At this stage, the mainstream off-ramp method is to generate lane-changing motives at appropriate places and use several lane-changing behaviors to realize the off-ramp process. However, since the lane-changing behavior itself cannot be adjusted according to the urgency of the off-ramp, the success rate of this method is low when driving off the expressway, and th...

AI Technical Summary

Problems solved by technology

However, since the lane-changing behavior itself cannot be adjusted according to the urgency of the off-ramp, the success rate of this method is low when driving off the expressway, and the required preparation distance is long, resulting in a decrease in the efficiency of driverless cars.

On the other hand, due to the limited perception range of unmanned vehicles and the driver's behavior on the highway is full of uncertainty, it is difficult to estimate the impact of the simple enumeration of lane-changing rules on the success rate of off-ramps, and cannot cover all environmental states; The results generated by pure learning methods are difficult to control, which will affect the safety and stability of vehicle driving

Images