Backend learnable decoder to facilitate autonomous vehicle operation

1 . A backend computing system comprising:

one or more processors; and

a memory storing instructions that, when executed by the one or more processors, cause the backend computing system to:

communicate with a fleet of vehicles that operate throughout a road network, including receiving, from at least a first vehicle of the fleet, over one or more networks, compressed sensor data from a database of the first vehicle;

execute a set of learnable decoders on the compressed sensor data to decompress the compressed sensor data and to perform a set of tasks, including one or more scene reconstruction tasks that increase the spatial dimensionality of a scene represented by the compressed sensor data;

generate or update an autonomy map for the road network based on the set of tasks performed by the set of learnable decoders; and

communicate, over one or more networks, the autonomy map to each vehicle of the fleet, the autonomy map facilitating autonomous driving by each vehicle throughout the road network.

2 . The backend computing system of claim 1 , wherein the set of tasks of the set of learnable decoders comprise (i) one or more scene understanding tasks, or (ii) one or more object detection tasks.

3 . The backend computing system of claim 2 , wherein the set of learnable decoders comprise a series of neural network layers on the backend computing system to perform at least one of (i) the one or more scene reconstruction tasks, (ii) the one or more scene understanding tasks, or (iii) the one or more object detection tasks.

4 . The backend computing system of claim 1 , wherein the one or more learnable decoders perform the one or more scene reconstruction tasks by generating a three or four-dimensional scene from one or more two-dimensional images.

5 . The backend computing system of claim 4 , wherein the one or more learnable decoders perform the one or more scene reconstruction tasks by performing neural radiance (NeRF) reconstruction to generate the three or four-dimensional scene.

6 . The backend computing system of claim 1 , wherein the compressed sensor data includes a representation of sensor fusion-based data that includes a combination of image data, radar data, and LIDAR data.

7 . The backend computing system of claim 1 , wherein the set of learnable decoders include one or more learnable decoders to reconstruct or simulate a path traveled by the first vehicle.

8 . The backend computing system of claim 1 , wherein the first vehicle includes at least one of a consumer-driven vehicle or a specialized mapping vehicle.

9 . The backend computing system of claim 1 , wherein at least one learnable decoder from the set of learnable decoders performs the one or more scene reconstruction tasks to generate reconstructed scenes of the road network based on the compressed sensor data.

10 . A non-transitory computer readable medium storing instructions that, when executed by one or more processors of a backend computing system, cause the backend computing system to:

communicate with a fleet of vehicles that operate throughout a road network, including receiving, from at least a first vehicle of the fleet, over one or more networks, compressed sensor data from a database of one or more vehicles of the fleet;

execute a set of learnable decoders on the compressed sensor data to decompress the compressed sensor data and to perform a set of tasks, including one or more scene reconstruction tasks that increase the spatial dimensionality of a scene represented by the compressed sensor data;

generate or update an autonomy map for the road network based on the set of tasks performed by the set of learnable decoders; and

communicate, over one or more networks, the autonomy map to each vehicle of the fleet, the autonomy map facilitating autonomous driving by each vehicle throughout the road network.

11 . The non-transitory computer readable medium of claim 10 , wherein the set of tasks of the set of learnable decoders comprise (i) one or more scene understanding tasks, or (ii) one or more object detection tasks.

12 . The non-transitory computer readable medium of claim 11 , wherein the set of learnable decoders comprise a series of neural network layers executed by the backend computing system to perform at least one of (i) the one or more scene reconstruction tasks, (ii) the one or more scene understanding tasks, or (iii) the one or more object detection tasks.

13 . The non-transitory computer readable medium of claim 10 , wherein the one or more learnable decoders perform the one or more scene reconstruction tasks by generating a three or four-dimensional scene from one or more two-dimensional images.

14 . The non-transitory computer readable medium of claim 13 , wherein the one or more learnable decoders perform the one or more scene reconstruction tasks by performing neural radiance (NeRF) reconstruction to generate the three or four-dimensional scene.

15 . The non-transitory computer readable medium of claim 13 , wherein the set of learnable decoders include one or more learnable decoders to reconstruct or simulate a path traveled by the first vehicle.

16 . The non-transitory computer readable medium of claim 10 , wherein the compressed sensor data includes a representation of sensor fusion-based data that includes a combination of image data, radar data, and LIDAR data.

17 . The non-transitory computer readable medium of claim 10 , wherein the first vehicle includes at least one of a consumer-driven vehicle or a specialized mapping vehicle.

18 . The non-transitory computer readable medium of claim 10 , wherein at least one learnable decoder from the set of learnable decoders performs the one or more scene reconstruction tasks to generate reconstructed scenes of the road network based on the compressed sensor data.

19 . A computer-implemented method of decoding data, the method being performed by one or more processors and comprising:

communicating with a fleet of vehicles that operate throughout a road network, including, receiving, from at least a first vehicle of the fleet, over one or more networks, compressed sensor data from a database of one or more vehicles of the fleet; and

executing a set of learnable decoders on the compressed sensor data to decompress the compressed sensor data and to perform a set of tasks, including one or more scene reconstruction tasks that increase the spatial dimensionality of a scene represented by the compressed sensor data;

generating or updating an autonomy map for the road network based on the set of tasks performed by the set of learnable decoders; and

communicating, over one or more networks, the autonomy map to each vehicle of the fleet, the autonomy map facilitating autonomous driving by each vehicle throughout the road network.

20 . The computer-implemented method of claim 19 , wherein the set of tasks of the set of learnable decoders comprise (i) one or more scene understanding tasks, or (ii) one or more object detection tasks.