Back to results
US 12,589,774 B1Granted
Prediction errors classification
AssigneeZoox, Inc.
InventorsMahsa Ghafarianzadeh, Ruitao Yi
Life of this patent
Estimating the term…
Abstract
Techniques for determining and aggregating prediction error are described herein. A prediction may be determined for an object within a region based on first sensor data. A prediction error may be determined based on a difference between the prediction and second sensor data. The prediction error may be classified by one or more error types and aggregated by a same error type. The aggregated prediction error may be used to train a machine learned prediction model.
Claims20
A system comprising:
- one or more processors;
- and
- one or more computer-readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising:
- receiving first sensor data captured by a vehicle located within a region at a first time, the first sensor data indicating an object;
- determining, using a machine-learned model, a predicted state associated with the object, the predicted state comprising at least a pose of the object at a second time or a maneuver performed by the object;
- receiving second sensor data captured by the vehicle at the second time, the second sensor data indicating the object;
- determining, based at least in part on the second sensor data, a state of the object at the second time;
- determining a prediction error of the machine-learned model based at least in part on a difference between the predicted state and the state of the object at the second time;
- determining, based at least in part on determining the prediction error, one or more classifications, wherein the one or more classifications:
- comprise at least one of an object classification of the object, an environmental feature classification of an environmental feature in the region, or a maneuver classification of the maneuver performed by the object, and
- comprise an indication, distinct from the difference between the predicted state and the state of the object at the second time, of the respective object classification, environmental feature classification, or maneuver classification being represented in at least one of the first sensor data or the second sensor data;
- determining, based at least in part on the one or more classifications, to associate the prediction error with an error type, wherein the error type comprises at least one of the one or more classifications;
- storing, as prediction error data and based at least in part on the error type, the prediction error and an indication of the error type;
- and
- training, for use by an autonomous vehicle, the machine-learned model based at least in part on the error type and the prediction error data.
12 more elements
A method comprising:
- receiving first data captured at a first time, the first data indicating an object;
- determining, using a machine-learned model, a predicted state associated with the object;
- receiving second data captured at a second time, the second data indicating the object;
- determining, based at least in part on the second data, a state of the object at the second time;
- determining a prediction error of the machine-learned model based at least in part on a difference between the predicted state and the state of the object at the second time;
- determining, based at least in part on determining the prediction error, one or more types, wherein the one or more types comprise at least one of an object classification of the object, an environmental feature classification of an environmental feature in a region associated with at least one of the first data or the second data, or a maneuver performed by the object;
- determining to associate the prediction error with an error type based at least in part on at least one the first data or the second data, wherein the error type comprises at least one of the one or more types;
- and
- training, for use by an autonomous vehicle, the machine-learned model based at least in part on the error type.
6 more elements
One or more non-transitory computer-readable media storing instructions executable by one or more processors, wherein the instructions, when executed, cause the one or more processors to perform operations comprising:
- receiving first data captured at a first time, the first data indicating an object;
- determining, using a machine-learned model and the first data, a predicted state associated with the object;
- receiving second data captured at a second time, the second data indicating the object;
- determining, based at least in part on the second data, a state of the object at the second time;
- determining a prediction error based at least in part on a difference between the predicted state and the state of the object at the second time;
- determining, based at least in part on determining the prediction error, one or more classifications, wherein the one or more classifications:
- are based at least in part on at least one of the first data or the second data, and
- comprise at least one of an object classification of the object, an environmental feature classification of an environmental feature in a region associated with at least one of the first data or the second data, or a maneuver classification of a maneuver performed by the object;
- determining to associate the prediction error with an error type based at least in part on at least one the first data or the second data, wherein the error type comprises at least one of the one or more classifications;
- and
- training, for use by an autonomous vehicle, the machine-learned model based at least in part on the error type.
8 more elements
Dependencies are inferred from the claim text.