PARKING STRATEGY BASED ON DEEP REINFORCEMENT LEARNING

摘要

Provided are a parking method and system based on deep reinforcement learning, and same relate to the field of intelligent driving, and particularly to a parking strategy based on deep reinforcement learning. In the prior art, a traditional automatic parking system is based on a traditional path planning algorithm, which is poor in effect. According to the present technical solution, a parking planning route can be acquired according to a deep reinforcement learning algorithm, and a tuple is formed based on a vehicle observation state, a vehicle prediction action and a reward function. The parking planning method based on the tuple has the characteristic of extracting features based on product characteristics, so that fewer parameters are required. In addition, based on an objective function: (distance + steering + collision), a coefficient is not required to be adjusted; and in the technical solution, the features are extracted by using a deep reinforcement learning method, which has the beneficial technical effects of being fast in overall planning time, fast in response to the outside, etc.