| CPC B60W 60/00 (2020.02) [G06V 10/764 (2022.01); G06V 20/58 (2022.01); B60W 2050/0062 (2013.01); B60W 2554/402 (2020.02); B60W 2556/10 (2020.02)] | 11 Claims |
|
1. A mobile object control device comprising:
processing circuitry configured to:
acquire, as learning history data, driving history data obtained when a learning mobile object is operated in a collision risk-free environment;
perform learning for imitating driving of the learning mobile object in the collision risk-free environment using the acquired learning history data as training data and generating an imitation learning model;
acquire, as training history data, driving history data obtained when a mobile object for generating training data is operated in a same environment as the environment in which the learning history data has been acquired;
performing a process of automatically labeling the training data by:
estimating whether the training history data matches the learning history data using the acquired training history data as input to the generated imitation learning model;
classifying a collision risk into a plurality of collision types based on at least a vehicle travel direction at an intersection and an emerging direction of an obstacle;
generating a collision risk label and a collision risk-free label by associating the collision risk with corresponding collision types among the plurality of collision types;
in response to a matching degree between the training history data and the learning history data being less than or equal to a predetermined threshold, determining that the mobile object for generating training data is in a driving state with a collision risk, and assigning the collision risk label to the training data; and
in response to the matching degree being greater than the predetermined threshold, determining that the mobile object for generating training data is in a collision risk-free driving state, and assigning the collision risk-free label to the training data;
learn a model for inferring vehicle control parameters for controlling a vehicle using the labeled training data that includes the collision risk label or the collision risk-free label, on a basis of sensor information of a sensor mounted on the vehicle; and
control an accelerator, a brake, and a steering wheel of the vehicle using the inferred vehicle control parameters,
wherein the imitation learning model is a different machine learning model from the model for inferring the vehicle control parameters, and the input of the imitation learning model is used to automatically label the training data for the model for inferring the vehicle control parameters.
|