Semantic object clustering for autonomous vehicle decision making
The technology relates to controlling a vehicle in an autonomous driving mode. For example, sensor data identifying a plurality of objects may be received. Pairs of objects of the plurality of objects may be identified. For each identified pair of objects of the plurality of objects, a similarity value which indicates whether the objects of that identified pair of objects can be responded to by the vehicle as a group may be determined. The objects of one of the identified pairs of objects may be clustered together based on the similarity score. The vehicle may be controlled in the autonomous mode by responding to each object in the cluster in a same way.
1. A method of controlling a vehicle in an autonomous driving mode, the method comprising:
receiving, by one or more processors, sensor data identifying a plurality of objects;
identifying pairs of objects of the plurality of objects; and
for each identified pair of objects of the plurality of objects:
determining, by the one or more processors, a similarity measure based on one or more factors associated with that identified pair of objects, wherein one of the one or more factors relates to past and current motion of the objects of the one of the identified pairs of objects;
determining, by the one or more processors, whether the similarity measure meets a predetermined threshold;
when the similarity measure is determined to meet the predetermined threshold, clustering, by the one or more processors, the objects of that identified pair of objects to form a cluster; and
controlling, by the one or more processors, the vehicle in the autonomous driving mode by responding to each object in the cluster in a same way.
2. The method of claim 1 , wherein one of the one or more factors relates to a distance between the objects of the one of the identified pairs of objects.
3. The method of claim 1 , wherein one of the one or more factors relates to object types of the objects of the one of the identified pairs of objects.
4. The method of claim 1 , wherein one of the one or more factors relates to whether one object of each identified pair of objects appears to be following another object of that identified pair of objects.
5. The method of claim 1 , wherein one of the one or more factors relates to whether the objects of each identified pair of objects are identified as belonging to a predetermined semantic group.
6. The method of claim 1 , further comprising, merging the cluster together with a second cluster based on whether the cluster and the second cluster include common objects, and wherein responding to each object in the cluster includes responding to each object in the merged cluster and second cluster.
7. The method of claim 6 , wherein the common objects are traffic cones.
8. The method of claim 1 , wherein the objects of one of the identified pairs of objects are both vehicles.
9. The method of claim 1 , wherein the objects of one of the identified pairs of objects are both parked vehicles.
10. The method of claim 1 , wherein the objects of the one of the identified pairs of objects are both stacked vehicles.
11. The method of claim 1 , wherein the objects of the one of the identified pairs of objects are both pedestrians.
12. The method of claim 1 , wherein the objects of the one of the identified pairs of objects are both bicyclists.
13. The method of claim 1 , wherein the objects of the one of the identified pairs of objects are both traffic cones.
14. The method of claim 1 , further comprising, reevaluating the cluster when new sensor data for the objects of the one of the identified pairs of objects is received.
15. The method of claim 1 , wherein identifying the pairs of objects of the plurality of objects includes:
identifying pairs of static objects of the plurality of objects; and
identifying pairs of moving objects of the plurality of objects.
16. A method of controlling a vehicle in an autonomous driving mode, the method comprising:
receiving, by one or more processors, sensor data identifying a plurality of objects;
identifying pairs of objects of the plurality of objects; and
for each identified pair of objects of the plurality of objects:
determining, by the one or more processors, a similarity measure based on one or more factors associated with that identified pair of objects, wherein one of the one or more factors relates to a similarity between predicted future motion of the objects of the one of the identified pairs of objects;
determining, by the one or more processors, whether the similarity measure meets a predetermined threshold;
when the similarity measure is determined to meet the predetermined threshold, clustering, by the one or more processors, the objects of that identified pair of objects to form a cluster; and
controlling, by the one or more processors, the vehicle in the autonomous driving mode by responding to each object in the cluster in a same way.
17. A method of controlling a vehicle in an autonomous driving mode, the method comprising:
receiving, by one or more processors, sensor data identifying a plurality of objects;
identifying pairs of objects of the plurality of objects; and
for each identified pair of objects of the plurality of objects:
determining, by the one or more processors, a similarity measure based on one or more factors associated with that identified pair of objects, wherein one of the one or more factors relates to a relative location of the objects of the one of the identified pairs of objects to a feature in an environment of the vehicle;
determining, by the one or more processors, whether the similarity measure meets a predetermined threshold;
when the similarity measure is determined to meet the predetermined threshold, clustering, by the one or more processors, the objects of that identified pair of objects to form a cluster; and
controlling, by the one or more processors, the vehicle in the autonomous driving mode by responding to each object in the cluster in a same way.
18. The method of claim 17 , wherein the feature is a crosswalk.
19. The method of claim 17 , wherein the feature is a bicycle lane.