US 12,269,512 B2
Method for training a machine learning algorithm for predicting an intent parameter for an object on a terrain
Boris Konstantinovich Yangel, Nalchik (RU); and Maksim Ilich Stebelev, Moscow (RU)
Assigned to Y.E. Hub Armenia LLC, Yerevan (AM)
Filed by YANDEX SELF DRIVING GROUP LLC, Moscow (RU)
Filed on Jan. 20, 2022, as Appl. No. 17/580,220.
Claims priority of application No. RU2021116056 (RU), filed on Jun. 3, 2021.
Prior Publication US 2022/0388547 A1, Dec. 8, 2022
Int. Cl. B60W 60/00 (2020.01); B60R 1/27 (2022.01); G06V 10/764 (2022.01); G06V 20/58 (2022.01)
CPC B60W 60/00276 (2020.02) [B60R 1/27 (2022.01); G06V 10/764 (2022.01); G06V 20/58 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A method of training a machine learning algorithm for predicting an intent parameter for an object on a terrain, a self-driving vehicle traveling on the terrain in proximity to the object, the method being executable by a server, the method comprising:
accessing, by the server, operational data associated with a training vehicle, the operational data having been previously stored in a storage, the operational data including data indicative of:
a training terrain, states of the training vehicle on the training terrain, and states of a training object on the training terrain;
for a target moment in time in the operational data:
retrieving, by the server from the operational data, data indicative of:
the training terrain at the target moment in time,
a target state of the training vehicle corresponding to the target moment in time, and a future state of the training vehicle after the target moment in time,
a target state of the training object corresponding to the target moment in time, and a future state of the training object after the target moment in time;
generating, by the server, a training set for training the machine learning algorithm, the training set having an input and an assessor-less label,
the input including the data indicative of the training terrain and of the target states of the training vehicle and the training object,
the assessor-less label being based on the data indicative of the future states of the training vehicle and the training object, the assessor-less label being indicative of an intent of the training object on the training terrain at the target moment in time; and
training, by the server, the machine learning algorithm based on the training set for predicting, during an in-use phase:
the intent parameter indicative of an intent of the object on the terrain during an in-use target moment in time based on data indicative of (i) the terrain at the in-use target moment in time, (ii) an in-use state of the self-driving vehicle at the in-use target moment in time, and (iii) an in-use state of the object;
triggering operation of the self-driving vehicle at least in part based on the intent parameter.