Approaches for encoding environmental information

1. A computer-implemented method comprising:

accessing, by a computing system, a plurality of schema-based encodings providing a structured representation of an environment, wherein the plurality of schema-based encodings are generated based at least in part on sensor data captured by one or more sensors associated with one or more vehicles while navigating the environment;

clustering, by the computing system, each schema-based encoding of the plurality of schema-based encodings into one or more clusters of schema-based encodings, wherein the clustering involves determining a similarity of one or more schema-based encodings of the plurality of schema-based encodings;

determining, by the computing system, at least one scenario associated with the environment based at least in part on the one or more clusters of schema-based encodings; and

subsequent to determining the at least one scenario:

determining, by the computing system, an exposure rate for the at least one scenario by evaluating the one or more clusters of schema-based encodings associated with the at least one scenario, wherein the exposure rate for the at least one scenario represents a frequency at which the at least one scenario, based at least in part on the similarity of the one or more schema-based encodings, was experienced by the one or more vehicles while navigating the environment.

2. The computer-implemented method of claim 1 , wherein a schema-based encoding of the environment for a period of time identifies one or more agents that were detected by a vehicle within the environment during the period of time, respective motion information for each of the one or more agents, information indicating whether an agent may potentially interact with the vehicle during the period of time, and metadata describing the environment.

3. The computer-implemented method of claim 1 , wherein clustering each schema-based encoding of the plurality of schema-based encodings further comprises:

generating, by the computing system, respective feature vector representations for each of the plurality of schema-based encodings; and

clustering, by the computing system, the feature vector representations based at least in part on similarity of the feature vector representations.

4. The computer-implemented method of claim 3 , wherein the clustering the feature vector representations based at least in part on similarity of the feature vector representations further comprises:

determining, by the computing system, that schema-based encodings included in a first cluster are associated with a first scenario family; and

determining, by the computing system, that schema-based encodings included in a second cluster are associated with a second scenario family.

5. The computer-implemented method of claim 4 , further comprising:

determining, by the computing system, that schema-based encodings included in a first sub-cluster of the first cluster are associated with a first scenario in the first scenario family; and

determining, by the computing system, that schema-based encodings included in a second sub-cluster of the first cluster are associated with a second scenario in the first scenario family.

6. The computer-implemented method of claim 1 , further comprising:

determining, by the computing system, a label for a first cluster in the one or more clusters of schema-based encodings; and

assigning, by the computing system, the label to unlabeled schema-based encodings included in the first cluster.

7. The computer-implemented method of claim 6 , wherein the label identifies at least one family of scenarios represented by schema-based encodings included in the first cluster.

8. The computer-implemented method of claim 6 , wherein the label identifies at least one scenario represented by schema-based encodings included in the first cluster.

9. The computer-implemented method of claim 1 , further comprising:

training, by the computing system, a machine learning model based at least in part on the labeled schema-based encodings included in the first cluster, wherein the machine learning model is capable of receiving a schema-based encoding of a navigated environment as input and outputting scenarios describing the navigated environment upon evaluating the inputted schema-based encoding.

10. The computer-implemented method of claim 1 , wherein the at least one scenario was experienced by the one or more vehicles while navigating the environment at different points in time.

11. A system comprising:

at least one processor; and

a memory storing instructions that, when executed by the at least one processor, cause the system to perform:

accessing a plurality of schema-based encodings providing a structured representation of an environment, wherein the plurality of schema-based encodings are generated based at least in part on sensor data captured by one or more sensors associated with one or more vehicles while navigating the environment;

clustering each schema-based encoding of the plurality of schema-based encodings into one or more clusters of schema-based encodings, wherein the clustering involves determining a similarity of one or more schema-based encodings of the plurality of schema-based encodings;

determining at least one scenario associated with the environment based at least in part on the one or more clusters of schema-based encodings; and

subsequent to determining the at least one scenario:

determining an exposure rate for the at least one scenario by evaluating the one or more clusters of schema-based encodings associated with the at least one scenario, wherein the exposure rate for the at least one scenario represents a frequency at which the at least one scenario, based at least in part on the similarity of the one or more schema-based encodings, was experienced by the one or more vehicles while navigating the environment.

12. The system of claim 11 , wherein a schema-based encoding of the environment for a period of time identifies one or more agents that were detected by a vehicle within the environment during the period of time, respective motion information for each of the one or more agents, information indicating whether an agent may potentially interact with the vehicle during the period of time, and metadata describing the environment.

13. The system of claim 11 , wherein clustering each schema-based encoding of the plurality of schema-based encodings further comprises:

generating respective feature vector representations for each of the plurality of schema-based encodings; and

clustering the feature vector representations based at least in part on similarity of the feature vector representations.

14. The system of claim 11 , wherein the clustering the feature vector representations based at least in part on similarity of the feature vector representations further causes the system to perform:

determining that schema-based encodings included in a first cluster are associated with a first scenario family; and

determining that schema-based encodings included in a second cluster are associated with a second scenario family.

15. The system of claim 14 , further comprising:

determining that schema-based encodings included in a first sub-cluster of the first cluster are associated with a first scenario in the first scenario family; and

determining that schema-based encodings included in a second sub-cluster of the first cluster are associated with a second scenario in the first scenario family.

16. A non-transitory computer-readable storage medium including instructions that, when executed by at least one processor of a computing system, cause the computing system to perform:

accessing a plurality of schema-based encodings providing a structured representation of an environment, wherein the plurality of schema-based encodings are generated based at least in part on sensor data captured by one or more sensors associated with one or more vehicles while navigating the environment;

clustering each schema-based encoding of the plurality of schema-based encodings into one or more clusters of schema-based encodings, wherein the clustering involves determining a similarity of one or more schema-based encodings of the plurality of schema-based encodings;

identifying at least one scenario associated with the environment based at least in part on the one or more clusters of schema-based encodings; and

subsequent to identifying the at least one scenario:

determining an exposure rate for the at least one scenario by evaluating the one or more clusters of schema-based encodings associated with the least one scenario, wherein the exposure rate for the at least one scenario represents a frequency at which the at least one scenario, based at least in part on the similarity of the one or more schema-based encodings, was experienced by the one or more vehicles while navigating the environment.

17. The non-transitory computer-readable storage medium of claim 16 , wherein a schema-based encoding of the environment for a period of time identifies one or more agents that were detected by a vehicle within the environment during the period of time, respective motion information for each of the one or more agents, information indicating whether an agent may potentially interact with the vehicle during the period of time, and metadata describing the environment.

18. The non-transitory computer-readable storage medium of claim 11 , wherein clustering each schema-based encoding of the plurality of schema-based encodings further comprises:

generating respective feature vector representations for each of the plurality of schema-based encodings; and

clustering the feature vector representations based at least in part on similarity of the feature vector representations.

19. The non-transitory computer-readable storage medium of claim 11 , wherein the clustering the feature vector representations based at least in part on similarity of the feature vector representations further causes the computing system to perform:

determining that schema-based encodings included in a first cluster are associated with a first scenario family; and

determining that schema-based encodings included in a second cluster are associated with a second scenario family.

20. The non-transitory computer-readable storage medium of claim 14 , further comprising:

determining that schema-based encodings included in a first sub-cluster of the first cluster are associated with a first scenario in the first scenario family; and

determining that schema-based encodings included in a second sub-cluster of the first cluster are associated with a second scenario in the first scenario family.