Semantic place recognition and localization

1. A computer-implemented method comprising:

receiving a sensor-based representation of a portion of a property, wherein the sensor-based representation of the portion of the property is derived from observations of the portion of the property that were obtained using one or more sensors of a robot while the robot was located at an unknown location of the property;

identifying, based at least on the sensor-based representation of the portion of the property, one or more objects that the sensor-based representation indicates as being located within the portion of the property;

determining, based at least on the one or more objects that the sensor-based representation indicates as being located within the portion of the property, a semantic zone type for the portion of the property;

accessing a model of the property that identifies (i) a plurality of semantic zones of the property, and (ii) for each semantic zone of the property, a corresponding semantic zone type for the semantic zone of the property and a corresponding location for the semantic zone of the property; and

selecting, from among the plurality of semantic zones of the property identified by the model, a particular semantic zone of the property where the robot was likely located when the observations of the portion of the property were obtained using the one or more sensors of the robot, wherein selecting the particular semantic zone comprises identifying a match between the determined semantic zone type for the portion of the property and the corresponding semantic zone type that is indicated by the model of the property for the particular semantic zone; and

setting, as a current location of the robot at the property, a particular location within the particular semantic zone of the property where the robot was likely located when the observations of the portion of the property were obtained using the one or more sensors of the robot.

2. The computer-implemented method of claim 1 , wherein the observations of the portion of the property comprise one or more images of the portion of the property.

3. The computer-implemented method of claim 1 , wherein the observations of the portion of the property comprise a plurality of LIDAR measurements obtained from the portion of the property.

4. The computer-implemented method of claim 1 , wherein determining the semantic zone type for the portion of the property comprises:

determining that the one or more objects that the sensor-based representation indicates as being located within the portion of the property is insufficient to determine a semantic zone type for the portion of the property;

receiving a second sensor-based representation of a different portion of the property, wherein the second sensor-based representation of the different portion of the property is derived from observations of the different portion of the property that were obtained using the one or more sensors of the robot;

identifying, based at least on the second sensor-based representation of the different portion of the property, one or more additional objects that the second sensor-based representation indicates as being located within the different portion of the property; and

determining, based at least on (i) the one or more objects that the sensor-based representation of the portion of the property indicates as being located within the portion of the property and (ii) the one or more additional objects that the second sensor-based representation indicates as being located within the different portion of the property, the semantic zone type for the portion of the property.

5. The computer-implemented method of claim 1 , wherein selecting the particular semantic zone of the property comprises:

identifying, using the model of the property, that the determined semantic zone type for the portion of the property matches the corresponding semantic zone type for multiple ones of the plurality of semantic zones of the property;

determining one or more characteristics of the portion of the property based on the sensor-based representation of the portion of the property;

comparing the one or more characteristics of the portion of the property to the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property that were identified as having a semantic zone type that matches the determined semantic zone type for the portion of the property; and

determining, based at least on a result of the comparison, that the robot was likely located in the particular semantic zone of the property when the observations of the portion of the property were obtained using the one or more sensors of the robot.

6. The computer-implemented method of claim 5 wherein the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property comprise indications of objects that are registered as being located in corresponding ones of the semantic zones of the property, and comparing the one or more characteristics of the portion of the property to the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property comprises comparing objects identified in the portion of the property to the objects that have been registered as being located in the corresponding ones of the semantic zones of the property.

7. The computer-implemented method of claim 5 , wherein the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property comprise corresponding three-dimensional representations of each of the multiple ones of the plurality of semantic zones of the property.

8. The computer-implemented method of claim 7 , wherein selecting the particular semantic zone of the property comprises:

determining, based at least on the sensor-based representation of the portion of the property, a three-dimensional representation of the portion of the property; and

comparing the three-dimensional representation of the portion of the property to the corresponding three-dimensional representations of the multiple ones of the plurality of semantic zones of the property.

9. A system comprising:

one or more processors configured to execute computer program instructions; and

one or more non-transitory computer-storage media encoded with computer programs that, when executed by the processor, cause the system to perform operations comprising:

receiving a sensor-based representation of a portion of a property, wherein the sensor-based representation of the portion of the property is derived from observations of the portion of the property that were obtained using one or more sensors of a robot while the robot was located at a particular position an unknown location of the property;

identifying, based at least on the sensor-based representation of the portion of the property, one or more objects that the sensor-based representation indicates as being located within the portion of the property;

determining, based at least on the one or more objects that the sensor-based representation indicates as being located within the portion of the property, a semantic zone type for the portion of the property;

accessing a model of the property that identifies (i) a plurality of semantic zones of the property, and (ii) for each semantic zone of the property, a corresponding semantic zone type for the semantic zone of the property and a corresponding location for the semantic zone of the property; and

selecting, from among the plurality of semantic zones of the property identified by the model, a particular semantic zone of the property where the robot was likely located when the observations of the portion of the property were obtained using the one or more sensors of the robot, wherein selecting the particular semantic zone comprises identifying a match between the determined semantic zone type for the portion of the property and the corresponding semantic zone type that is indicated by the model of the property for the particular semantic zone; and

setting, as a current location of the robot at the property, a particular location within the particular semantic zone of the property where the robot was likely located when the observations of the portion of the property were obtained using the one or more sensors of the robot.

10. The system of claim 9 , wherein the observations of the portion of the property comprise one or more images of the portion of the property.

11. The system of claim 9 , wherein the observations of the portion of the property comprise a plurality of LIDAR measurements obtained from the portion of the property.

12. The system of claim 9 , wherein determining the semantic zone type for the portion of the property comprises:

determining that the one or more objects that the sensor-based representation indicates as being located within the portion of the property is insufficient to determine a semantic zone type for the portion of the property;

receiving a second sensor-based representation of a different portion of the property, wherein the second sensor-based representation of the different portion of the property is derived from observations of the different portion of the property that were obtained using the one or more sensors of the robot;

identifying, based at least on the second sensor-based representation of the different portion of the property, one or more additional objects that the second sensor-based representation indicates as being located within the different portion of the property; and

determining, based at least on (i) the one or more objects that the sensor-based representation of the portion of the property indicates as being located within the portion of the property and (ii) the one or more additional objects that the second sensor-based representation indicates as being located within the different portion of the property, the semantic zone type for the portion of the property.

13. The system of claim 9 , wherein selecting the particular semantic zone of the property comprises:

identifying, using the model of the property, that the determined semantic zone type for the portion of the property matches the corresponding semantic zone type for multiple ones of the plurality of semantic zones of the property;

determining one or more characteristics of the portion of the property based on the sensor-based representation of the portion of the property;

comparing the one or more characteristics of the portion of the property to corresponding characteristics of the multiple ones of the plurality of semantic zones of the property that were identified as having a semantic zone type that matches the determined semantic zone type for the portion of the property; and determining, based at least on a result of the comparison, that the robot was likely located in the particular semantic zone of the property when the observations of the portion of the property were obtained using the one or more sensors of the robot.

14. The system of claim 9 , wherein the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property comprise corresponding three-dimensional representations of each of the multiple ones of the plurality of semantic zones of the property.

15. One or more non-transitory computer-readable devices storing software comprising instructions executable by one or more computers which, upon such execution, cause the one or more computers to perform operations comprising:

receiving a sensor-based representation of a portion of a property, wherein the sensor-based representation of the portion of the property is derived from observations of the portion of the property that were obtained using one or more sensors of a robot while the robot was located at an unknown location of the property;

identifying, based at least on the sensor-based representation of the portion of the property, one or more objects that the sensor-based representation indicates as being located within the portion of the property;

determining, based at least on the one or more objects that the sensor-based representation indicates as being located within the portion of the property, a semantic zone type for the portion of the property;

accessing a model of the property that identifies (i) a plurality of semantic zones of the property, and (ii) for each semantic zone of the property, a corresponding semantic zone type for the semantic zone of the property and a corresponding location for the semantic zone of the property; and

selecting, from among the plurality of semantic zones of the property identified by the model, a particular semantic zone of the property where the robot was likely located when the observations of the portion of the property were obtained using the one or more sensors of the robot, wherein selecting the particular semantic zone comprises identifying a match between the determined semantic zone type for the portion of the property and the corresponding semantic zone type that is indicated by the model of the property for the particular semantic zone; and

setting, as a current location of the robot at the property, a particular location within the particular semantic zone of the property where the robot was likely located when the observations of the portion of the property were obtained using the one or more sensors of the robot.

16. The one or more non-transitory computer-readable devices of claim 15 , wherein the observations of the portion of the property comprise one or more images of the portion of the property.

17. The one or more non-transitory computer-readable devices of claim 15 , wherein the observations of the portion of the property comprise a plurality of LIDAR measurements obtained from the portion of the property.

18. The one or more non-transitory computer-readable devices of claim 15 , wherein determining the semantic zone type for the portion of the property comprises:

determining that the one or more objects that the sensor-based representation indicates as being located within the portion of the property is insufficient to determine a semantic zone type for the portion of the property;

receiving a second sensor-based representation of a different portion of the property, wherein the second sensor-based representation of the different portion of the property is derived from observations of the different portion of the property that were obtained using the one or more sensors of the robot;

identifying, based at least on the second sensor-based representation of the different portion of the property, one or more additional objects that the second sensor-based representation indicates as being located within the different portion of the property; and

determining, based at least on (i) the one or more objects that the sensor-based representation of the portion of the property indicates as being located within the portion of the property and (ii) the one or more additional objects that the second sensor-based representation indicates as being located within the different portion of the property, the semantic zone type for the portion of the property.

19. The one or more non-transitory computer-readable devices of claim 15 , wherein selecting the particular semantic zone of the property comprises:

identifying, using the model of the property, that the determined semantic zone type for the portion of the property matches the corresponding semantic zone type for multiple ones of the plurality of semantic zones of the property;

determining one or more characteristics of the portion of the property based on the sensor-based representation of the portion of the property;

comparing the one or more characteristics of the portion of the property to corresponding characteristics of the multiple ones of the plurality of semantic zones of the property that were identified as having a semantic zone type that matches the determined semantic zone type for the portion of the property; and

determining, based at least on a result of the comparison, that the robot was likely located in the particular semantic zone of the property when the observations of the portion of the property were obtained using the one or more sensors of the robot.

20. The one or more non-transitory computer-readable devices of claim 19 , wherein the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property comprise indications of objects that are registered as being located in corresponding ones of the semantic zones of the property, and comparing the one or more characteristics of the portion of the property to the corresponding characteristics of the multiple ones of the plurality of semantic zones of the property comprises comparing objects identified in the portion of the property to the objects that have been registered as being located in the corresponding ones of the semantic zones of the property.