Environment Adaptive Import Method for Intelligent Driving Vehicles in Urban Environment

Abstract

The invention discloses an intelligent driving vehicle environment adaptive merging method under an urban environment. The method comprises steps: an initial state vector is extracted; an action variable is calculated according to a greedy strategy, a merging scene is updated while a merging action is executed, if the action variable adopts a random action, a merging gap and a merging action are selected with a uniform probability, if an intelligent method is adopted, candidate gaps comprise a front vehicle, a following vehicle and a merging vehicle, the maximum action value functions of all candidate gaps are compared, the maximum value function is selected, the gap and the action corresponding to the maximum value are picked out, and a target merging gap and an intelligent merging actionare returned; the state vector at a next moment is sensed; a reward value is calculated according to the environmental feedback information; the initial state vector, the action variable, the state vector at the next moment and the reward value are saved to a sample set, and after enough samples are obtained, evaluation and improvement are carried out according to an LSQ method; and the above steps are repeated until merging succeeds. The sample set and the learning time are lower than a Q learning algorithm, and the success rate is high.

Description

technical field [0001] The method relates to an environment-adaptive import method that comprehensively considers the selection of the target gap and the desired timing of the import in a complex urban environment. Background technique [0002] As the future development trend of transportation, unmanned vehicles have great potential in solving traffic safety and road congestion problems. As the "brain" of unmanned vehicles, the decision-making system reflects its intelligence level. Improving the generalization and adaptability of the decision-making system in complex urban environments is crucial for the development of unmanned vehicles that can actually drive on the road. However, traditional rule-based learning unmanned vehicles can only adapt to a single driving environment, and cannot face complex and changeable real scenes, and the decisions made may not meet the requirements of robustness and flexibility. Expressway merging in an urban environment requires safe and e...

AI Technical Summary

Problems solved by technology

Shalev-Shwartz discussed a safe reinforcement learning method, which divides the policy network into two parts and learns driving safety and comfort separately, but the effectiveness of the model is only verified in a simple simulation environment

[0005] The above method only considers a few indicators when considering the import, and cannot simulate the automatic import driving experience of human driving

Images