System and method for a scenario-based event trigger
A method for scenario-based event triggers is described. The method includes generating, by a first machine-learning (ML) model, feature vectors encoding driving scenarios surrounding an ego vehicle. The method also includes detecting, by a second machine-learning (ML) model, a unique driving scenario outside of pre-programmed event triggers corresponding to one of the feature vectors encoding driving scenarios surrounding the ego vehicle. The method further includes triggering uploading of the unique driving scenario outside of pre-programmed event triggers to a central scenario-based event control server.
1 . A method for scenario-based event triggers, comprising:
generating, by a first machine-learning (ML) model, feature vectors encoding driving scenarios surrounding an ego vehicle;
detecting, by a second machine-learning (ML) model, a unique driving scenario outside of pre-programmed event triggers corresponding to one of the feature vectors encoding driving scenarios surrounding the ego vehicle;
uploading, by the ego vehicle, recorded scenario data corresponding to the unique driving scenario outside of the pre-programmed event triggers to a central scenario-based event control server;
triggering, by the ego vehicle, a fleet of vehicles to record and upload scenario data corresponding to each fleet detected similar event related to the unique driving scenario outside of the pre-programmed event triggers to the central scenario-based event control server;
providing, by the central scenario-based event control server, additional instructions on what event should or should not trigger the fleet upload including the priority of the triggers which are used to determine which logs are deleted in the event where the non-uploaded data exceed the vehicle data storage;
training a new model according to the upload of the recorded scenario data corresponding to the unique driving scenario and each fleet detected similar event related to the unique driving scenario outside of the pre-programmed event triggers; and
operating the ego vehicle with the new model to safely maneuver through the unique driving scenario outside of the pre-programmed event triggers.
2 . The method of claim 1 , in which generating the feature vectors encoding driving scenarios comprises:
feeding sensor data captured by the ego vehicle to the first ML model to generate vectors of hash values encoding driving scenarios surrounding the ego vehicle; and
determining whether a current driving scenario surrounding the ego vehicle corresponds to a rule-based event trigger.
3 . The method of claim 1 , further comprising recording of the unique driving scenario outside of pre-programmed event triggers.
4 . The method of claim 1 , in which the detecting a unique driving scenario comprises:
receiving an output of a perception model of the ego vehicle;
identifying a detection result of the perception model for which the perception model indicates a lack of confidence; and
triggering an event in response to the identifying the detection result of the perception model for which the perception model indicates the lack of confidence.
5 . The method of claim 1 , further comprising receiving, from the central scenario-based event control server, additional instructions regarding trigger events according to a geolocation, model number, and/or a sensor package version of the ego vehicle.
6 . The method of claim 5 , in which the additional instructions comprises a length of an event recorded, a type of sensor to record, an event trigger priority, and/or an identification of logs deleted when an event data exceeds a vehicle data storage.
7 . The method of claim 1 , further comprising categorizing and detecting driving scenarios identified as error pruning, complex and/or unique.
8 . A non-transitory computer-readable medium having program code recorded thereon for scenario-based event triggers, the program code being executed by a processor and comprising:
program code to generate, by a first machine-learning (ML) model, feature vectors encoding driving scenarios surrounding an ego vehicle;
program code to detect, by a second machine-learning (ML) model, a unique driving scenario outside of pre-programmed event triggers corresponding to one of the feature vectors encoding driving scenarios surrounding the ego vehicle;
program code to upload, by the ego vehicle, recorded scenario data corresponding to the unique driving scenario outside of the pre-programmed event triggers to a central scenario-based event control server;
program code to trigger, by the ego vehicle, a fleet of vehicles to record and upload scenario data corresponding to each fleet detected similar event related to the unique driving scenario outside of pre-programmed event triggers to the central scenario-based event control server;
program code to provide, by the central scenario-based event control server, additional instructions on what event should or should not trigger the fleet upload including the priority of the triggers which are used to determine which logs are deleted in the event where the non-uploaded data exceed the vehicle data storage;
program code to train a new model according to the upload of the recorded scenario data corresponding to the unique driving scenario and each fleet detected similar event related to the unique driving scenario outside of the pre-programmed event triggers; and
program code to operate the ego vehicle using the new model to safely maneuver through the unique driving scenario outside of the pre-programmed event triggers.
9 . The non-transitory computer-readable medium of claim 8 , in which the program code to generate the feature vectors encoding driving scenarios comprises:
program code to feed sensor data captured by the ego vehicle to the first ML model to generate vectors of hash values encoding driving scenarios surrounding the ego vehicle; and
program code to determine whether a current driving scenario surrounding the ego vehicle corresponds to a rule-based event trigger.
10 . The non-transitory computer-readable medium of claim 8 , further comprising program code to record the unique driving scenario outside of pre-programmed event triggers.
11 . The non-transitory computer-readable medium of claim 8 , in which the program code to detect a unique driving scenario comprises:
program code to receive an output of a perception model of the ego vehicle;
program code to identify a detection result of the perception model for which the perception model indicates a lack of confidence; and
program code to trigger the event in response to the identifying the detection result of the perception model for which the perception model indicates the lack of confidence.
12 . The non-transitory computer-readable medium of claim 8 , further comprising program code to receive, from the central scenario-based event control server, additional instructions regarding trigger events according to a geolocation, model number, and/or a sensor package version of the ego vehicle.
13 . The non-transitory computer-readable medium of claim 12 , in which the additional instructions comprises a length of the event recorded, a type of sensor to record, an event trigger priority, and/or an identification of logs deleted when an event data exceeds a vehicle data storage.
14 . The non-transitory computer-readable medium of claim 8 , further comprising program code to categorize and detect driving scenarios identified as error pruning, complex and/or unique.
15 . A system for scenario-based event triggers, comprising:
a feature vector generator to generate feature vectors encoding driving scenarios surrounding an ego vehicle using a first machine-learning (ML) model;
a scenario-based event trigger to detect a unique driving scenario outside of pre-programmed event triggers corresponding to one of the feature vectors encoding driving scenarios surrounding the ego vehicle using a second machine-learning (ML) model and to upload, by the ego vehicle, recorded scenario data corresponding to the unique driving scenario outside of the pre-programmed event triggers to a central scenario-based event control server; and
a driving-scenario upload trigger module to trigger, by the ego vehicle, a fleet of vehicles to record and upload scenario data corresponding to each fleet detected similar event related to the unique driving scenario outside of pre-programmed event triggers to the central scenario-based event control server, to provide, by the central scenario-based event control server, additional instructions on what event should or should not trigger the fleet upload including the priority of the triggers which are used to determine which logs are deleted in the event where the non-uploaded data exceed the vehicle data storage, and to train a new model according to the upload of the recorded scenario data corresponding to the unique driving scenario and each fleet detected similar event related to the unique driving scenario outside of the pre-programmed event triggers and to operate the ego vehicle using the new model to safely maneuver through the unique driving scenario outside of the pre-programmed event triggers.
16 . The system of claim 15 , in which the feature vector generator is further to feed sensor data captured by the ego vehicle to the first ML model to generate vectors of hash values encoding driving scenarios surrounding the ego vehicle, and to determine whether a current driving scenario surrounding the ego vehicle corresponds to a rule-based event trigger.
17 . The system of claim 15 , further comprising the central scenario-based event control server to provide additional instructions regarding trigger events according to a geolocation, model number, and/or a sensor package version of the ego vehicle.
18 . The system of claim 15 , in which the driving-scenario upload trigger module is further configured to categorize and detect driving scenarios identified as error pruning, complex and/or unique.