Method and apparatus for performing multi-agent sensor fusion in cooperative driving based on reinforcement learning

摘要

PROBLEM TO BE SOLVED: To provide a method and an apparatus for learning a sensor fusion network for sensor fusion of an autonomous vehicle that performs cooperative driving. A learning device (100) provides a running image including an autonomously running vehicle, m cooperative driving vehicles, and a second virtual vehicle and sensor state information for n sensors of the m cooperative driving vehicles. Inputting to the fusion network to generate a sensor fusion probability value which is a probability value of transmitting sensor values of n sensors, and inputting a road running image to the detection network to input a second virtual vehicle and a pedestrian. , A lane is detected, peripheral object information is output, the sensor value and peripheral object information are input to the traveling network to generate a moving direction probability value, the autonomous traveling vehicle is operated, and traffic condition information is displayed. Acquiring and generating rewards to learn the sensor fusion network. [Selection diagram] Fig. 3