Adversarial learning of driving behavior
AssigneeToyota Research Institute, Inc.
InventorOluwatobi O. Olabiyi
Life of this patent
Estimating the term…
Abstract
Embodiments described herein disclose methods and systems for adversarial learning in autonomous vehicle path modeling. The systems and methods collect states of the vehicle in the environment to predict a path. The predicted path is compared for variance from an actual path. The variance between the paths, in light of other data, is used to modify the driving models, to create more accurate representations of expert driving in autonomous vehicle path generation.
Claims22
An adversarial learning system for path comparison in autonomous learning, comprising:
- one or more processors;
- and
- a memory communicably coupled to the one or more processors and storing instructions that when executed by the one or more processors cause the one or more processors to:
- generate a predicted path between an origin and a destination for a vehicle using an autonomous driving model;
- obtain an actual path driven by an operator of the vehicle between the origin and the destination;
- calculate a variance factor using the predicted path and the actual path;
- modify the autonomous driving model using the variance factor, the variance factor being applied only to a portion of the autonomous driving model pertaining to at least one of:
- specific decisions, specific decision types, or specific time frames within the predicted path, a magnitude of the variance factor being altered based on at least one of:
- mechanical limits of the vehicle, road conditions, known vehicle responsiveness limitations;
- and
- using the modified autonomous driving model to autonomously drive the vehicle.
A non-transitory computer-readable medium for path comparison in autonomous learning and storing instructions that when executed by one or more processors cause the one or more processors to:
- generate a predicted path between an origin and a destination for a vehicle using an autonomous driving model;
- obtain an actual path driven by an operator of the vehicle between the origin and the destination;
- calculate a variance factor using the predicted path and the actual path;
- modify the autonomous driving model using the variance factor, the variance factor being applied only to a portion of the autonomous driving model pertaining to at least one of:
- specific decisions, specific decision types, or specific time frames within the predicted path, a magnitude of the variance factor being altered based on at least one of:
- mechanical limits of the vehicle, road conditions, known vehicle responsiveness limitations;
- and
- using the modified autonomous driving model to autonomously drive the vehicle.
A method for path comparison in autonomous learning, comprising:
- generating a predicted path between an origin and a destination for a vehicle using an autonomous driving model;
- obtaining an actual path driven by an operator of the vehicle between the origin and the destination;
- calculating a variance factor using the predicted path and the actual path;
- and
- modifying the autonomous driving model using the variance factor, the variance factor being applied only to a portion of the autonomous driving model pertaining to at least one of:
- specific decisions, specific decision types, or specific time frames within the predicted path, a magnitude of the variance factor being altered based on at least one of:
- mechanical limits of the vehicle, road conditions, known vehicle responsiveness limitations;
- and
- using the modified autonomous driving model to autonomously drive the vehicle.
Dependencies are inferred from the claim text.