Database generation including predicted scenarios
Techniques for generating a database containing predicted driving scenarios are discussed herein. A database management component may receive sensor data captured by sensors of vehicle(s) (e.g., driving log data) based on previous driving trips within various physical driving environments. The database management component may cluster driving scenarios observed from the sensor data based on the driving scenario's feature similarity. In some examples, the database management component may determine prediction information (e.g., actual trajectory from log data or trajectory generated by machine-learning model) to associate with each cluster. Accordingly, each cluster may include an encoded representation (e.g., key) as well as corresponding prediction information (e.g., value). In some examples, the database management component may store some or all key-value pairs in a database accessible by one or more vehicles while such vehicles navigate an environment.
1 . A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing computer-executable instructions that, when executed, cause the one or more processors to perform operations comprising:
receiving log data associated with an environment, wherein the log data is representative of a set of driving scenarios that were experienced at a previous time;
inputting the log data into a machine-learning model;
receiving, from the machine-learning model, a set of encoded data representative of the set of driving scenarios;
determining, based at least in part on the set of encoded data, a cluster of encoded data of the set of encoded data representative of driving scenarios of the set of driving scenarios;
determining, based at least in part on the cluster of encoded data associated with the driving scenarios, prediction information associated with an object in the cluster of the encoded data associated with the driving scenarios;
storing, in a database, the cluster and the prediction information; and
transmitting the database to an autonomous vehicle configured to be controlled based at least in part on the database.
2 . The system of claim 1 , wherein the cluster includes a key, the key being determined based at least in part on determining an average value of the cluster.
3 . The system of claim 1 , wherein a driving scenario is based at least in part on at least one of:
a road geometry associated with the set of driving scenarios,
a first number of lanes associated with the set of driving scenarios,
a first type of lane associated with the set of driving scenarios,
a speed limit associated with the set of driving scenarios,
a second number of objects associated with the set of driving scenarios,
a second type of object associated with the set of driving scenarios,
environmental data associated with the set of driving scenarios, or
a feature of an object associated with the set of driving scenarios.
4 . The system of claim 1 , wherein the prediction information comprises at least one of:
a previously observed trajectory of the object,
a trajectory generated based at least in part on the log data, or
a reference to a second cluster.
5 . The system of claim 4 , wherein the trajectory is generated based at least in part on:
inputting the set of driving scenarios into a machine-learned model;
receiving, from the machine-learned model, a set of predicted states of the environment; and
determining, based at least in part on the set of predicted states, the trajectory.
6 . 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 log data representative of a set of driving scenarios that were experienced at a previous time;
determining, based at least in part on the log data, a set of encoded data representative of the set of driving scenarios;
determining a cluster of encoded data of the set of encoded data representative of driving scenarios of the set of driving scenarios;
determining, based at least in part on the cluster, a representative scenario of the set of driving scenarios;
determining a prediction associated with the representative scenario;
storing, in a database, an association between the representative scenario and the prediction; and
transmitting, to an autonomous vehicle, the database, the autonomous vehicle configured to be controlled based at least in part on the database.
7 . The one or more non transitory computer readable media of claim 6 , wherein determining the set of encoded data comprises:
inputting the set of driving scenarios into a machine learned model; and
receiving, from the machine learned model, a set of encodings, an encoding of the set of encodings representative of a driving scenario of the set of driving scenarios.
8 . The one or more non transitory computer readable media of claim 7 , wherein determining the representative scenario further comprises:
clustering the set of encodings into a plurality of clusters; and
determining, based at least in part on a subset of encodings associated with the cluster of the plurality of clusters, the representative scenario.
9 . The one or more non transitory computer readable media of claim 8 , wherein the cluster of the plurality of clusters includes a key, the key being determined based at least in part on determining an average value of the cluster.
10 . The one or more non transitory computer readable media of claim 8 , wherein determining the representative scenario is based at least in part on determining an average of two or more trajectories associated with the cluster of the plurality of clusters.
11 . The one or more non transitory computer readable media of claim 6 , wherein the set of driving scenarios are defined based at least in part on one or more of:
a road geometry associated with the set of driving scenarios,
a first number of lanes associated with the set of driving scenarios,
a first type of lane associated with the set of driving scenarios,
a speed limit associated with the set of driving scenarios,
a second number of objects associated with the set of driving scenarios,
a second type of object associated with the set of driving scenarios,
environmental data associated with the set of driving scenarios, or
a feature of an object associated with the set of driving scenarios.
12 . The one or more non transitory computer readable media of claim 6 , wherein the prediction comprises at least one of:
a previously observed trajectory of an object associated with the representative scenario,
a trajectory generated based at least in part on the log data, or
a reference to a second cluster.
13 . The one or more non transitory computer readable media of claim 12 , wherein the trajectory is generated based at least in part on:
inputting the set of driving scenarios into a machine-learned model;
receiving, from the machine-learned model, a set of predicted states of an environment; and
determining, based at least in part on the set of predicted states, the trajectory.
14 . A method comprising:
receiving log data representative of a set of driving scenarios that were experienced at a previous time;
determining, based at least in part on the log data, a set of encoded data representative of the set of driving scenarios;
determining a cluster of encoded data of the set of encoded data representative of driving scenarios of the set of driving scenarios;
determining, based at least in part on the cluster, a representative scenario of the set of driving scenarios;
determining a prediction associated with the representative scenario;
storing, in a database, an association between the representative scenario and the prediction; and
transmitting, to an autonomous vehicle, the database, the autonomous vehicle configured to be controlled based at least in part on the database.
15 . The method of claim 14 , wherein determining the set of encoded data comprises:
inputting the set of driving scenarios into a machine learned model; and
receiving, from the machine learned model, a set of encodings, an encoding of the set of encodings representative of a driving scenario of the set of driving scenarios.
16 . The method of claim 15 , wherein determining the representative scenario further comprises:
clustering the set of encodings into a plurality of clusters; and
determining, based at least in part on a subset of encodings associated with the cluster of the plurality of clusters, the representative scenario.
17 . The method of claim 16 , wherein the cluster of the plurality of clusters includes a key, the key being determined based at least in part on determining an average value of the cluster.
18 . The method of claim 16 , wherein determining the representative scenario is based at least in part on determining an average of two or more trajectories associated with the cluster of the plurality of clusters.
19 . The method of claim 14 , wherein the set of driving scenarios are defined based at least in part on one or more of:
a road geometry associated with the set of driving scenarios,
a first number of lanes associated with the set of driving scenarios,
a first type of lane associated with the set of driving scenarios,
a speed limit associated with the set of driving scenarios,
a second number of objects associated with the set of driving scenarios,
a second type of object associated with the set of driving scenarios,
environmental data associated with the set of driving scenarios, or
a feature of an object associated with the set of driving scenarios.
20 . The method of claim 14 , wherein the prediction comprises at least one of:
a previously observed trajectory of an object associated with the representative scenario,
a trajectory generated based at least in part on the log data, or
a reference to a second cluster.