Zoox, Inc.

Dropbox, Inc.

Andraz Kavalar

Gerrit Bagschik

Kai Zhenyu Wang

Gowtham Garimella

Andrew Scott Crego

Michael Haggblade

Jongmin Baek

Daniel R Horn

Austin Sung

David Lichtenberg

Anne-Claire Elisabeth Marie Le Henaff

Michael Loh

Shi Feng

Techniques are discussed herein for executing and evaluating simulations and/or emulations based on parameterized scenarios, such as driving scenarios used to analyze and evaluate the responses of autonomous vehicle controllers. A surrogate model may be determined by simulating points in a simulation parameter space with a simulator to determine the surrogate model while capturing the main features of the simulation model. Additionally, relevant areas of the parameter space, such as, areas containing parameters associated with events occurring infrequently during simulations, may be quickly identified using Bayesian optimization techniques. An iterative evaluation of an acquisition function may delineate a contour associated with the region of interest, and the surrogate model may be evaluated to determine a probability of returning a result corresponding to a parameter being within the contour. The probabilities may be further analyzed to determine scenario/vehicle performance metrics access the scenario parameter space.

Probability calculation of encountering scenarios using emulator models

Validating a component of an autonomous vehicle may comprise determining, via simulation, a likelihood that operation of the component will result in an adverse event. Such simulations may be based on log data developed from real world driving events to, for example, accurately model a likelihood that a scenario will occur during real-world driving. Because adverse events may be exceedingly rare, the techniques may include modifying a probability distribution associated the likelihood that a scenario is simulated, determining a metric associated with an adverse event (e.g., a likelihood that operating the vehicle or updating a component thereof will result in an adverse event), and applying a correction to the metric based on the modification to the probability distribution.

Vehicle component validation using adverse event simulation

Techniques for top-down scene discrimination are discussed. A system receives scene data associated with an environment proximate a vehicle. The scene data is input to a convolutional neural network (CNN) discriminator trained using a generator and a classification of the output of the CNN discriminator. The CNN discriminator generates an indication of whether the scene data is a generated scene or a captured scene. If the scene data is data generated scene, the system generates a caution notification indicating that a current environmental situation is different from any previous situations. Additionally, the caution notification is communicated to at least one of a vehicle system or a remote vehicle monitoring system.

Top-down scene discrimination

Techniques for top-down scene generation are discussed. A generator component may receive multi-dimensional input data associated with an environment. The generator component may generate, based at least in part on the multi-dimensional input data, a generated top-down scene. A discriminator component receives the generated top-down scene and a real top-down scene. The discriminator component generates binary classification data indicating whether an individual scene in the scene data is classified as generated or classified as real. The binary classification data is provided as a loss to the generator component and the discriminator component.

Top-down scene generation

Techniques are discussed herein for determining truncated simulation regions within a parameter space of simulation scenarios, such as driving scenarios used to analyze and evaluate the responses of autonomous vehicle controllers. Using non-sampling-based parameter selection techniques, parameterized scenarios may be executed as simulations to determine the truncated simulation region. Sampling-based parameter selection techniques may be used to determine additional parameterized scenarios, which may be compared to the truncated simulation region. Parameterized scenarios within the truncated simulation region may be executed as simulations and scenarios outside of the truncated simulation region may be excluded, and the aggregated results may be analyzed to determine scenario/vehicle performance metrics across the scenario parameter space.

Scenario simulation execution within a truncated parameter space

Systems and methods for dynamic and automatic data storage scheme switching in a distributed data storage system. A machine learning-based policy for computing probable future content item access patterns based on historical content item access patterns is employed to dynamically and automatically switch the storage of content items (e.g., files, digital data, photos, text, audio, video, streaming content, cloud documents, etc.) between different data storage schemes. The different data storage schemes may have different data storage cost and different data access cost characteristics. For example, the different data storage schemes may encompass different types of data storage devices, different data compression schemes, and/or different data redundancy schemes.

Growing community of inventors

San Francisco, CA, United States of America

Andraz Kavalar