Allstate Insurance Company

Leland Stanford Junior University

Juan Carlos Aragon

Regina Madigan

Sunil Kumar Chintakindi

Derek Phillips

Mykel Kochenderfer

Methods, computer-readable media, software, and apparatuses may determine whether a human driver, or an autonomous driving system, should be in control of a vehicle in response to a detection of an unexpected event. A decision may be made to pass control from the autonomous driving system to the human driver, if it is determined that the human driver can handle the unexpected event more safely than the autonomous vehicle.

Safe hand-off between human driver and autonomous driving system

Aspects of the disclosure relate to enhanced processing systems for providing dynamic driving metric outputs using improved machine learning methods. A computing platform may receive sensor data from vehicle sensors. The computing platform may generate a pattern deviation output corresponding to an output of a sensor data analysis model, an actual outcome associated with a lowest TTC value, and driving actions that occurred over a prediction horizon corresponding to the pattern deviation output. The computing platform may cluster the pattern deviation outputs to maximize a ratio of inter-cluster variance to intra-cluster variance. The computing platform may train a long short term memory (LSTM) for each cluster, and may verify consistency of the pattern deviation outputs in the respective clusters. After verifying the consistency of the pattern deviation outputs in each cluster, the computing platform may modify the sensor data analysis model to reflect pattern deviation outputs associated with verified consistency.

Adaptable on-deployment learning platform for driver analysis output generation

Aspects of the disclosure relate to a dynamic distance estimation output platform that utilizes improved computer vision and perspective transformation techniques to determine vehicle proximities from video footage. A computing platform may receive, from a visible light camera located in a first vehicle, a video output showing a second vehicle that is in front of the first vehicle. The computing platform may determine a longitudinal distance between the first vehicle and the second vehicle by determining an orthogonal distance between a center-of-projection corresponding to the visible light camera, and an intersection of a backside plane of the second vehicle and ground below the second vehicle. The computing platform may send, to an autonomous vehicle control system, a distance estimation output corresponding to the longitudinal distance, which may cause the autonomous vehicle control system to perform vehicle control actions.

Dynamic distance estimation output generation based on monocular video

Aspects of the disclosure relate to dynamic driving metric output platforms that utilize improved computer vision methods to determine vehicle metrics from video footage. A computing platform may receive video footage from a vehicle camera. The computing platform may determine that a reference marker in the video footage has reached a beginning and an end of a road marker based on brightness transitions, and may insert time stamps into the video accordingly. Based on the time stamps, the computing platform may determine an amount of time during which the reference marker covered the road marking. Based on a known length of the road marking and the amount of time during which the reference marker covered the road marking, the computing platform may determine a vehicle speed. The computing platform may generate driving metric output information, based on the vehicle speed, which may be displayed by an accident analysis platform. Based on known dimensions of pavement markings the computing platform may obtain the parameters of the camera (e.g., focal length, camera height above ground plane and camera tilt angle) used to generate the video footage and use the camera parameters to determine the distance between the camera and any object in the video footage.

Dynamic driving metric output generation using computer vision methods

Aspects of the disclosure relate to an autonomous vehicle evaluation system that performs continuous evaluation of the actions, strategies, preferences, margins, and responses of an autonomous driving control system. A computing platform may receive sensor data from one or more autonomous vehicle sensors, manufacturer computing platform, or V2X computing platform. Based on this sensor data, the computing platform may determine one or more driving patterns. Based on a primary context corresponding to the one or more driving patterns, the computing platform may group the one or more driving patterns. The computing platform may determine a driving pattern degradation output indicating degradation corresponding to the one or more grouped driving patterns, and the computing platform may send the driving pattern degradation output to an autonomous driving system, which may cause the autonomous driving system to take corrective action accordingly.

Growing community of inventors

Redwood City, CA, United States of America

Juan Carlos Aragon