Method and device

Abstract

The invention provides a method for learning a sensor fusion network for sensor fusion of an autonomous vehicle performing a cooperative driving. The method includes the following steps: learning equipment (a) inputs (i) a driving image including the autonomous vehicle, m cooperatively-driving vehicles, and second virtual vehicles and (ii) sensor status information on n sensors in the m cooperatively-driving vehicles into the sensor fusion network, to generate sensor fusion probabilities of sensor values of the n sensors being transmitted, (b) inputs a road-driving video into a detection network, to detect the second virtual vehicles, pedestrians, and lanes and output nearby object information, and inputs sensor values and the nearby object information into a drive network, to generate moving direction probabilities and drive the autonomous vehicle and (c) acquires traffic condition information, generates a reward, and learns the sensor fusion network.

Description

technical field [0001] The present invention relates to a method and apparatus for autonomous driving vehicles, virtual driving, etc.; more particularly, to a method and apparatus for performing multi-agent sensor fusion based on reinforcement learning in autonomous driving. Background technique [0002] The vehicle industry has recently shifted to an era of environmentally friendly and advanced vehicles incorporating IT technology. In addition to the development of vehicle technology, smart vehicles are also being commercialized, in which technologies such as accident prevention, accident avoidance, collision safety, convenience improvement, vehicle informatization, and autonomous driving technologies are applied. [0003] This intelligent vehicle provides comfort functions by providing assistive technology and voice recognition for driver distraction and unskilled operation, thereby reducing accidents caused by driver negligence, and is also expected to reduce time, fuel w...

AI Technical Summary

Problems solved by technology

[0007] Therefore, there is a problem that even sensors that are unnecessary according to the current driving environment are continuously operated, so that the self-driving vehicle consumes a large amount of power

[0008] To solve this problem, conventional methods employ sensor fusion to use only the best sensors needed to detect the driving environment

[0009] In addition, recently, although autonomous vehicles can share information with each other through V2V communication (i.e., vehicle-to-vehicle communication), a large amount of communication resources are consumed due to sending and receiving a large amount of sensor data, and due to all the sending and receiving sensor The data must be processed, and the consumption of computing resources is also very large

Images