Procedurally generated three-dimensional environment for use in autonomous vehicle simulations

1. A computer-implemented method for generating a synthetic map, comprising:

collecting, by an autonomous vehicle traversing a real-world environment, sensor data corresponding with a three-dimensional (3D) space in the real-world scene and first outputs of an autonomous vehicle stack as the autonomous vehicle traverses the real-world environment;

generating a 3D mesh based on the sensor data;

generating one or more synthetic 3D objects based on the 3D mesh and the sensor data;

generating a 3D synthetic environment comprising the one or more synthetic 3D objects, wherein the 3D synthetic environment is generated based on the 3D mesh;

simulating navigation of a simulated autonomous vehicle, which is operating using the autonomous vehicle stack, through at least a portion of the 3D map to generate a second outputs of the autonomous vehicle stack; and

comparing the first outputs of the autonomous vehicle stack to the second outputs of the autonomous vehicle stack to determine a validity of the 3D synthetic environment.

2. The computer-implemented method of claim 1 , wherein the sensor data comprises Light Detection and Ranging (LiDAR) sensor data collected by the autonomous vehicle.

3. The computer-implemented method of claim 1 , wherein the sensor data comprises Light Detection and Ranging (LiDAR) sensor data associated with one or more roadways.

4. The computer-implemented method of claim 1 , wherein generating the 3D mesh based on the sensor data, further comprises:

seeding the 3D mesh using the sensor data.

5. The computer-implemented method of claim 1 , wherein the 3D mesh comprises a footprint for one or more buildings.

6. The computer-implemented method of claim 1 , wherein the one or more comprises one or more: buildings, cross walks, traffic lights, stop signs, or sidewalks.

7. A system comprising:

one or more processors; and

a computer-readable medium comprising instructions stored therein, which when executed by the one or more processors, cause the one or more processors to perform operations comprising:

collecting, by a first vehicle traversing a real-world environment, first sensor data corresponding with a three-dimensional (3D) space in the real-world scene;

parsing the first sensor data into a tile format for downstream processing;

generating a set of map tiles based on the first sensor data in the tile format and a set of labels, wherein the set of map tiles include a combination of raster map and semantic map data;

collecting, by a second vehicle, which is an autonomous vehicle, traversing the real-world environment, second sensor data corresponding with the three-dimensional (3D) space in the real-world scene;

performing one or more of localization, maneuvering, and navigation using the second sensor data and the set of map tiles to determine the validity of the second sensor data based on the set of map tiles;

generating, following validation, a 3D mesh based on the set of map tiles;

generating one or more synthetic 3D objects based on the 3D mesh and the first sensor data; and

generating a 3D synthetic map comprising the one or more synthetic 3D objects, wherein the 3D synthetic environment is generated based on the 3D mesh.

8. The system method of claim 7 , wherein the first sensor data comprises Light Detection and Ranging (LiDAR) sensor data collected by an autonomous vehicle (AV).

9. The system method of claim 7 , wherein the first sensor data comprises Light Detection and Ranging (LiDAR) sensor data associated with one or more roadways.

10. The system method of claim 7 , wherein generating the 3D mesh based on the set of map tiles, further comprises:

seeding the 3D mesh using the set of map tiles.

11. The system method of claim 7 , wherein the 3D mesh comprises a foot-print for one or more buildings.

12. The system method of claim 7 , wherein the one or more synthetic objects comprises one or more: buildings, cross walks, traffic lights, stop signs, or sidewalks.

13. A non-transitory computer-readable storage medium comprising instructions stored therein, which when executed by one or more processors, cause the processors to perform operations comprising:

collecting, by an autonomous vehicle traversing a real-world environment, sensor data corresponding with a three-dimensional (3D) space in the real-world scene and first outputs of an autonomous vehicle stack as the autonomous vehicle traverses the real-world environment;

generating a 3D mesh based on the sensor data;

generating one or more synthetic 3D objects based on the 3D mesh and the sensor data;

generating a 3D synthetic environment comprising the one or more synthetic 3D objects, wherein the 3D synthetic environment is generated based on the 3D mesh;

simulating navigation of a simulated autonomous vehicle, which is operating using the autonomous vehicle stack, through at least a portion of the 3D map to generate a second outputs of the autonomous vehicle stack; and

comparing the first outputs of the autonomous vehicle stack to the second outputs of the autonomous vehicle stack to determine a validity of the 3D synthetic environment.

14. The non-transitory computer-readable storage medium of claim 13 , wherein the sensor data comprises Light Detection and Ranging (LiDAR) sensor data collected by the autonomous vehicle.

15. The non-transitory computer-readable storage medium of claim 13 , wherein the sensor data comprises Light Detection and Ranging (LiDAR) sensor data associated with one or more roadways.

16. The non-transitory computer-readable storage medium of claim 13 , wherein generating the 3D mesh based on the sensor data, further comprises:

seeding the 3D mesh using the sensor data.

17. The non-transitory computer-readable storage medium of claim 13 , wherein the 3D mesh comprises a foot-print for one or more buildings.