A learning method and a learning device for performing a trajectory-based behavior analysis on a recent travel route to determine whether to change a vehicle mode from a manual travel mode to an autonomous travel mode

摘要

PROBLEM TO BE SOLVED: To provide a learning method for calculating a collision probability used for judging whether or not a traveling mode switching of a vehicle capable of autonomous traveling is appropriate by analyzing a recent traveling route of a driver. .. SOLUTION: (a) When a learning device acquires a state vector and a trajectory vector, (i) a process of generating a state feature map with a state network, and (ii) a trajectory feature map with a trajectory network. (B) causing the learning device to calculate a predicted collision probability, which is a probability that an accident is predicted to occur, with the safety network; and (c) the learning. A method in which a device learns at least some of the parameters by generating a loss with reference to the predicted collision probability and the acquired GT (Ground Truth) collision probability with a loss layer. [Selection diagram] Fig. 3