Back to results
US 12,606,206 B2Granted
Filtering autonomous driving simulation scenarios
AssigneeGM Cruise Holdings LLC
InventorsAlessandro Iyra, Vivek Rangaswamy
Life of this patent
Estimating the term…
Abstract
A computer system for filtering autonomous driving scenarios for simulations includes circuitry and a graphical processing unit (GPU). The circuitry receives data encoding a first autonomous driving scenario, and generates a first embedding of the first autonomous driving scenario based on a model for transforming autonomous driving scenario data to an embedding. A GPU compares the first embedding to a test bank of embeddings representing other autonomous driving scenarios. If the first embedding has at least a threshold similarity to other embeddings in the test bank, the first autonomous driving scenario is discarded from a simulation pipeline for simulating autonomous driving scenarios.
Claims14
A method comprising:
- receiving data encoding a first autonomous driving scenario, wherein the first autonomous driving scenario indicates a set of objects in an environment, including locations and types of each object in the set of objects;
- compressing the data encoding the first autonomous driving scenario into a rasterized encoding by converting the data encoding the first autonomous driving scenario from one reference frame to a different reference frame corresponding to a perspective of an autonomous vehicle, the rasterized encoding comprising vectors storing locations of the set of objects within a coordinate system;
- generating a first embedding of the first autonomous driving scenario based on a model for transforming autonomous driving scenario data to an embedding, the first embedding generated from the rasterized encoding and having a lower dimensionality than the data encoding the first autonomous driving scenario, wherein the model is trained to output a value indicating a similarity of a set of autonomous driving scenarios and produce an embedding for each autonomous driving scenario input to the model;
- inputting the first embedding and a second embedding of a second autonomous driving scenario to a graphical processing unit (GPU), wherein the second autonomous driving scenario corresponds to a set of scenarios selected for use in a simulation pipeline for simulating autonomous driving scenarios for testing and validating algorithms, machine learning models, and neural networks used by a real-world autonomous vehicle and the first autonomous driving scenario corresponds to a proposed scenario for use in the simulation pipeline;
- comparing, by the GPU, the first embedding of the first autonomous driving scenario to the second embedding of the second autonomous driving scenario to determine a similarity between the first autonomous driving scenario and the second autonomous driving scenario;
- in response to determining that the first embedding has less than a threshold similarity to the second embedding, adding the first autonomous driving scenario to the set of scenarios selected for use in the simulation pipeline;
- performing, by the simulation pipeline, a first simulation based on the first autonomous driving scenario to produce a first simulation outcome;
- performing, by the simulation pipeline, a second simulation based on the second autonomous driving scenario to produce a second simulation outcome;
- determining an outcome similarity score between the first simulation outcome and the second simulation outcome;
- and
- in response to determining that a difference between the outcome similarity score and the similarity between the first autonomous driving scenario and the second autonomous driving scenario satisfies a refinement threshold, refining the model.
8 more elements
A system for filtering autonomous driving simulation scenarios, the system comprising:
- circuitry to:
- receive data encoding a first autonomous driving scenario, wherein the first autonomous driving scenario indicates a set of objects in an environment, including locations and types of each object in the set of objects;
- compress the data encoding the first autonomous driving scenario into a rasterized encoding by converting the data encoding the first autonomous driving scenario from one reference frame to a different reference frame corresponding to a perspective of an autonomous vehicle, the rasterized encoding comprising vectors storing locations of the set of objects within a coordinate system;
- generate a first embedding of the first autonomous driving scenario based on a model for transforming autonomous driving scenario data to an embedding, the first embedding generated from the rasterized encoding and having a lower dimensionality than the data encoding the first autonomous driving scenario, wherein the model is trained to output a value indicating a similarity of a set of autonomous driving scenarios and produce an embedding for each autonomous driving scenario input to the model;
- and
- input the first embedding and a second embedding of a second autonomous driving scenario to a graphical processing unit (GPU), wherein the second autonomous driving scenario corresponds to a set of scenarios selected for use in a simulation pipeline for simulating autonomous driving scenarios for testing and validating algorithms, machine learning models, and neural networks used by a real-world autonomous vehicle and the first autonomous driving scenario corresponds to a proposed scenario for use in the simulation pipeline;
- compare the first embedding of the first autonomous driving scenario to the second embedding of the second autonomous driving scenario to determine a similarity between the first autonomous driving scenario and the second autonomous driving scenario;
- and
- the circuitry further to:
- in response to determining that the first embedding has less than a threshold similarity to the second embedding, add the first autonomous driving scenario to the set of scenarios selected for use in the simulation pipeline;
- perform, by the simulation pipeline, a first simulation based on the first autonomous driving scenario to produce a first simulation outcome;
- perform, by the simulation pipeline, a second simulation based on the second autonomous driving scenario to produce a second simulation outcome;
- determine an outcome similarity score between the first simulation outcome and the second simulation outcome;
- and
- in response to determining that a difference between the outcome similarity score and the similarity between the first autonomous driving scenario and the second autonomous driving scenario satisfies a refinement threshold, refine the model.
12 more elements
A non-transitory computer-readable medium storing instructions for filtering autonomous driving simulations, the instructions, when executed by a processor, cause the processor to:
- receive data encoding a first autonomous driving scenario, wherein the first autonomous driving scenario indicates a set of objects in an environment, including locations and types of each object in the set of objects;
- compress the data encoding the first autonomous driving scenario into a rasterized encoding by converting the data encoding the first autonomous driving scenario from one reference frame to a different reference frame corresponding to a perspective of an autonomous vehicle, the rasterized encoding comprising vectors storing locations of the set of objects within a coordinate system;
- generate a first embedding of the first autonomous driving scenario based on a model for transforming autonomous driving scenario data to an embedding, the first embedding generated from the rasterized encoding and having a lower dimensionality than the data encoding the first autonomous driving scenario, wherein the model is trained to output a value indicating a similarity of a set of autonomous driving scenarios and produce an embedding for each autonomous driving scenario input to the model;
- input the first embedding and a second embedding of a second autonomous driving scenario to a graphical processing unit (GPU), wherein the second autonomous driving scenario corresponds to a set of scenarios selected for use in a simulation pipeline for simulating autonomous driving scenarios for testing and validating algorithms, machine learning models, and neural networks used by a real-world autonomous vehicle and the first autonomous driving scenario corresponds to a proposed scenario for use in the simulation pipeline;
- compare, by the GPU, the first embedding of the first autonomous driving scenario to the second embedding of the second autonomous driving scenario to determine a similarity between the first autonomous driving scenario and the second autonomous driving scenario;
- in response to determining that the first embedding has less than a threshold similarity to the second embedding, add the first autonomous driving scenario to the set of scenarios selected for use in the simulation pipeline;
- perform, by the simulation pipeline, a first simulation based on the first autonomous driving scenario to produce a first simulation outcome;
- perform, by the simulation pipeline, a second simulation based on the second autonomous driving scenario to produce a second simulation outcome;
- determine an outcome similarity score between the first simulation outcome and the second simulation outcome;
- and
- in response to determining that a difference between the outcome similarity score and the similarity between the first autonomous driving scenario and the second autonomous driving scenario satisfies a refinement threshold, refine the model.
8 more elements
Dependencies are inferred from the claim text.