Method and system for automatic driving data collection and closed-loop management

A system for automatic driving data collection and closed-loop management, applied in a vehicle, the system comprising:

• a data obtaining unit configured to obtain vehicle driving data;
• a data processor comprising:
• a data preprocessing module configured to preprocess the vehicle driving data;
• and
• a data filtering module, a pre-trained neural network being deployed on the data filtering module, and the data filtering module being configured to obtain incremental data by filtering, through a pre-trained neural network based on a predetermined filtering rule, the preprocessed vehicle driving data;
• and
• a memory having the incremental data stored thereon;
• and/or
• a transmitting module configured to transmit the incremental data to a cloud,
• wherein the vehicle driving data comprises visual sensed data, and the predetermined filtering rule comprises a comparison and filtering rule for a high-precision map and sensed data, a comparison and filtering rule for prediction data and sensed data, and a filtering rule for adjacent frame image difference,
• the comparison and filtering rule for the high-precision map and the sensed data comprises determining whether current identification data comprises a marked target road infrastructure in the high precision map to determine the incremental data;
• the comparison and filtering rule for the predication data and the sensed data comprises determining whether the predication data is same as the sensed data based on a predicted first position of an obstacle vehicle and an actually sensed second position of the obstacle vehicle to determine the incremental data;
• and
• the filtering rule for the adjacent frame image difference comprises determining whether a difference index between sensed images in adjacent frames is greater than a third threshold to determine the incremental data,
• or,
• wherein the vehicle driving data comprises vehicle control data, and the vehicle control data comprises current vehicle state data St and a current control action ahuman from a driver of the vehicle,
• said obtaining the incremental data by filtering, through the pre-trained neural network based on the predetermined filtering rule, the preprocessed vehicle driving data comprises:
• collecting a current control action ann output, based on the current vehicle state data St, by a pre-trained deep reinforcement learning neural network;
• calculating a difference between the current control action ann output by the deep reinforcement learning neural network and the current control action ahuman from the driver;
• determining whether the difference is greater than a fourth threshold;
• determining a rewarding and punishing function value r(t) of the deep reinforcement learning neural network based on the current vehicle state data St and the current control action ann output by the deep reinforcement learning neural network, in response to determining that the difference is greater than the fourth threshold;
• and
• determining a data packet as the incremental data, the data packet being constructed based on the current vehicle state data St, the current control action ann output by the deep reinforcement learning neural network, the current control action ahuman from the driver, the rewarding and punishing function value r(t), and vehicle state data St+1 at a next moment.