Automatically detecting unmapped drivable road surfaces for autonomous vehicles
Aspects of the disclosure relate to detecting unmapped drivable road surfaces. In one instance, sensor data captured by a sensor of an autonomous vehicle may be projected onto a grid having a plurality of cells. The plurality of cells may be classified by generating a label for each of the plurality of cells. Each label may identifies whether or not a corresponding cell contains a drivable surface. Ones of the plurality of cells may be clustered based on the labels to form a cluster of cells. An area of the cluster of cells may be compared to a map. Whether the area of the cluster of cells is an unmapped drivable road surface may be determined based on the comparison.
1. A method of detecting unmapped drivable road surfaces, the method comprising:
projecting, by one or more processors, sensor data captured by a sensor of an autonomous vehicle onto a grid having a plurality of cells, wherein the plurality of cells associated with the grid correspond to a plurality of cells associated with a pre-stored map of an area where the sensor data was captured;
classifying, by the one or more processors, the plurality of cells by generating a label for each of the plurality of cells, each label identifying whether or not a corresponding cell contains a drivable surface; and
determining, by the one or more processors, one or more areas of any of the plurality of cells as an unmapped drivable road surface based on the labels.
2. The method of claim 1 , wherein the sensor data includes LIDAR sensor data.
3. The method of claim 1 , wherein the sensor data includes camera images.
4. The method of claim 1 , wherein the classifying includes using a machine learning classifier to generate the labels.
5. The method of claim 1 , further comprising classifying, by the one or more processors, the plurality of cells by generating a second label for each of the plurality of cells, each second label identifying one or more driving directions for a corresponding cell, and wherein the determining is further based on the second labels.
6. The method of claim 1 , further comprising classifying, by the one or more processors, the plurality of cells by generating a second label for each of the plurality of cells, each second label identifying whether a corresponding cell is part of an intersection, and wherein the determining is further based on the second labels.
7. The method of claim 1 , further comprising classifying, by the one or more processors, the plurality of cells by generating a second label for each of the plurality of cells, each second label identifying whether a corresponding cell is part of a driveway, and wherein the determining is further based on the second labels.
8. The method of claim 1 , further comprising classifying, by the one or more processors, the plurality of cells by generating a second label for each of the plurality of cells, each second label identifying whether a corresponding cell is part of a crosswalk, and wherein the determining is further based on the second labels.
9. The method of claim 1 , further computing grouping adjacent cells to form a cluster based on the labels, and wherein the determining is further based on the grouping.
10. The method of claim 1 , further comprising determining a confidence for each of the labels, and wherein the classifying is further based on the confidences.
11. The method of claim 10 , further comprising grouping together non-adjacent cells to form a cluster based on the confidences, and wherein the determining is further based on the grouping.
12. The method of claim 11 , wherein the grouping includes ignoring one or more labels having confidence levels below a threshold when forming the cluster.
13. The method of claim 1 , further comprising includes determining an amount of overlap between the one or more areas and a region of the pre-stored map having a label identifying whether or not a corresponding cell contains a drivable surface, and wherein the determining is further based on the amount of overlap.
14. The method of claim 13 , wherein when the amount of overlap is not a complete overlap, the determining includes determined to be an unmapped drivable road surface.
15. The method of claim 13 , wherein when the amount of overlap is a complete overlap, the one or more areas are determined not to be an unmapped drivable road surface.
16. The method of claim 13 , wherein when the amount of overlap is a partial overlap, the method further comprises determining that the one or more areas correspond to a moved road feature.
17. The method of claim 1 , further comprising using the determination to control an autonomous vehicle in an autonomous driving mode.
18. The method of claim 17 , wherein controlling the autonomous vehicle using the determination includes changing a driving behavior of the autonomous vehicle.
19. The method of claim 17 , further comprising sending the determination to other vehicles having autonomous driving modes.
20. The method of claim 1 , further comprising using the determination as a signal that other unmapped drivable road surfaces may exist in an area proximate to any cells determined to be an unmapped drivable road surface.
21. The method of claim 1 , wherein a size and location of the plurality of cells of the grid correspond to a size and location of the plurality of cells associated with the pre-stored map.