Map change detection
The present technology provides systems, methods, and devices that can update aspects of a map as an autonomous vehicle navigates a route, and therefore avoids the need for dispatching a special purpose mapping vehicle for these updates. As the autonomous vehicle navigates the route, data captured by at least one sensor of an autonomous vehicle can indicate an inconsistency between pre-mapped from a high-resolution sensor system describing a location on a map, and current data describing a new feature of the location. The current data can be clustered together based on a threshold spatial closeness, where the clustering describes the new feature, and semantic labels of the pre-mapped data from the high-resolution sensor system can be updated based on the new feature described by the clustered current data.
1. A method comprising:
receiving, as an autonomous vehicle navigates a route, data indicating an inconsistency between pre-mapped data describing a location on a map, and current data describing a feature of the location, wherein the current data is captured by at least one sensor of the autonomous vehicle;
clustering the current data together based on associated intensity of the feature and a threshold spatial closeness of pixels within the feature, wherein the clustering describes the feature;
determining a structure of the feature based on an analysis of horizontal cross-sections in the current data, wherein the structure of the feature includes lane information of a road;
determining that a semantic label associated with the structure of the feature does not correspond to semantic labels associated with the feature in the pre-mapped data; and
updating the semantic labels of the pre-mapped data based on the feature described by the clustered current data to generate modified pre-mapped data.
2. The method of claim 1 , the method further comprising:
classifying a type of the feature in accordance with the analysis of the structure of the clustered current data; and
combining the modified pre-mapped data with revised semantic labels, the revised semantic labels having been generated based on the type of the feature.
3. The method of claim 1 , the method further comprising:
connecting a pixel of the feature with another pixel of the feature when the another pixel is within a threshold distance of the pixel;
clustering connected pixels into the feature when the connected pixels describe a shape of the feature; and
removing pixels that do not fall within the clustered connected pixels, wherein the map is revised based on iterating through the clustered pixels that represent inconsistencies with the pre-mapped data.
4. The method of claim 3 , wherein the connected pixels are clustered based on a determination that the shape of the feature is within a same direction as the road.
5. The method of claim 1 , wherein the current data describing the feature of the location is captured by at least one sensor capturing a change in the associated intensity, and wherein the threshold spatial closeness of high intensity areas are narrower than a threshold width, the high intensity areas being above a threshold intensity of low intensity areas.
6. The method of claim 1 , the method further comprising:
defining a map portion of the map by a boundary of location coordinates, wherein the pre-mapped data is an earlier version of the map portion, and the clustered current data is a later, revised version of the map portion; and
modifying the map in accordance with the revised version of the map portion when the revised version of the map portion is within a drivable location for the autonomous vehicle.
7. A computing system comprising:
at least one non-transitory computer readable medium comprising instructions stored thereon, wherein the instructions are effective to cause the computing system to:
receive, as an autonomous vehicle navigates a route, data indicating an inconsistency between pre-mapped data describing a location on a map, and current data describing a feature of the location, wherein the current data is captured by at least one sensor of the autonomous vehicle;
cluster the current data together based on associated intensity of the feature and a threshold spatial closeness of pixels within the feature, wherein the clustering describes the feature;
determine a structure of the feature based on an analysis of horizontal cross-sections in the current data, wherein the structure of the feature includes lane information of a road;
determine that a semantic label associated with the structure of the feature does not correspond to semantic labels associated with the feature in the pre-mapped data; and
update the semantic labels of the pre-mapped data based on the feature described by the clustered current data to generate modified pre-mapped data.
8. The computing system of claim 7 , wherein the instructions are effective to cause the computing system to further:
classify a type of the feature in accordance with the analysis of the structure of the clustered current data; and
revise the map by combining the modified pre-mapped data with revised semantic labels, the revised semantic labels having been generated based on the type of the feature.
9. The computing system of claim 7 , wherein the instructions are effective to cause the computing system to further:
connect a pixel of the feature with another pixel of the feature when the another pixel is within a threshold distance of the pixel;
cluster the connected pixels into the feature when the connected pixels describe a shape of the feature; and
remove pixels that do not fall within the clustered connected pixels, wherein the map is revised based on iterating through the clustered pixels that represent inconsistencies with the pre-mapped data.
10. The computing system of claim 9 , wherein the connected pixels are clustered based on a determination that the shape of the feature is within a same direction as the road.
11. The computing system of claim 7 , wherein the current data describing the feature of the location is captured by at least one sensor capturing a change in the associated intensity, and wherein the threshold spatial closeness of high intensity areas are narrower than a threshold width, the high intensity areas being above a threshold intensity of low intensity areas.
12. The computing system of claim 7 , wherein the instructions are effective to cause the computing system to further:
define a map portion of the map by a boundary of location coordinates, wherein the pre-mapped data is an earlier version of the map portion, and the clustered current data is a later, revised version of the map portion; and
modify the map in accordance with the revised version of the map portion when the revised version of the map portion is within a drivable location for the autonomous vehicle.
13. At least one non-transitory computer readable medium comprising instructions stored thereon, wherein the instructions are effective to cause an autonomous vehicle to:
receive, as the autonomous vehicle navigates a route, data indicating an inconsistency between pre-mapped data describing a location on a map, and current data describing a feature of the location, wherein the current data is captured by at least one sensor of the autonomous vehicle;
cluster the current data together based on associated intensity of the feature and a threshold spatial closeness of pixels within the feature, wherein the clustering describes the feature;
determine a structure of the feature based on an analysis of horizontal cross-sections in the current data, wherein the structure of the feature includes lane information of a road;
determine that a semantic label associated with the structure of the feature does not correspond to semantic labels associated with the feature in the pre-mapped data; and
update the semantic labels of the pre-mapped data based on the feature described by the clustered current data to generate modified pre-mapped data.
14. The at least one non-transitory computer readable medium of claim 13 , wherein the instructions are effective to cause the autonomous vehicle to:
classify a type of the feature in accordance with the analysis of the structure of the clustered current data; and
revise the map by combining the modified pre-mapped data with revised semantic labels, the revised semantic labels having been generated based on the type of the feature.
15. The at least one non-transitory computer readable medium of claim 13 , wherein the instructions are effective to cause the autonomous vehicle to:
connect a pixel of the feature with another pixel of the feature when the another pixel is within a threshold distance of the pixel;
cluster the connected pixels into the feature when the connected pixels describe a shape of the feature; and
remove pixels that do not fall within the clustered connected pixels, wherein the map is revised based on iterating through the clustered pixels that represent inconsistencies with the pre-mapped data.
16. The at least one non-transitory computer readable medium of claim 15 , wherein the connected pixels are clustered based on a determination that the shape of the feature is within a same direction as the road.
17. The at least one non-transitory computer readable medium of claim 13 , wherein the current data describing the feature of the location is captured by at least one sensor capturing a change in the associated intensity, and wherein the threshold spatial closeness of high intensity areas are narrower than a threshold width, the high intensity areas being above a threshold intensity of low intensity areas.