Systems and methods for vehicle-to-vehicle communications for improved autonomous vehicle operations

1. A computer-implemented method for vehicle-to-vehicle communications, the method comprising:

obtaining, by a first autonomous vehicle, sensor data associated with an environment of the first autonomous vehicle;

determining an intermediate environmental representation of at least a portion of the environment of the first autonomous vehicle based at least in part on the sensor data;

generating a compressed intermediate environmental representation by compressing the intermediate environmental representation of at least the portion of the environment of the first autonomous vehicle;

selecting a second autonomous vehicle to which to communicate the compressed intermediate environmental representation from among a plurality of autonomous vehicles based at least in part on an ability of the second autonomous vehicle to decompress the compressed intermediate environmental representation, wherein selecting the second autonomous vehicle comprises:

communicating with the second autonomous vehicle as the second autonomous vehicle enters a communication range of the first autonomous vehicle;

exchanging, with the second autonomous vehicle, data indicating that the second autonomous vehicle is able to decompress the compressed intermediate environmental representation; and

selecting the second autonomous vehicle based on the data indicating that the second autonomous vehicle is able to decompress the compressed intermediate environmental representation; and

communicating the compressed intermediate environmental representation to the second autonomous vehicle.

2. The computer-implemented method of claim 1 , wherein the sensor data comprises three-dimensional point cloud data, and wherein determining the intermediate environmental representation comprises:

generating voxelized sensor data by voxelizing the three-dimensional point cloud data;

inputting the voxelized sensor data into a machine-learned model, the machine-learned model configured to apply one or more convolutional layers to the voxelized sensor data; and

obtaining the intermediate environmental representation as an output of the machine-learned model.

3. The computer-implemented method of claim 1 , wherein the intermediate environmental representation comprises a feature map describing at least the portion of the environment of the first autonomous vehicle.

4. The computer-implemented method of claim 1 , further comprising:

selecting the second autonomous vehicle to which to communicate the compressed intermediate environmental representation from among a plurality of autonomous vehicles.

5. The computer-implemented method of claim 4 , wherein selecting the second autonomous vehicle to which to communicate the compressed intermediate environmental representation from among the plurality of autonomous vehicles comprises:

selecting, by the computing system, the second autonomous vehicle based at least in part on the communication range of the first autonomous vehicle.

6. The computer-implemented method of claim 1 , wherein the sensor data comprises LIDAR point cloud data.

7. The computer-implemented method of claim 1 , wherein the sensor data comprises a first type of sensor data and a second type of sensor data, wherein the first type of sensor data is associated with a first sensor modality, and the second type of sensor data is associated with a second sensor modality.

8. The computer-implemented method of claim 1 , wherein the sensor data comprises a first set of sensor data acquired by the first autonomous vehicle and a second set of sensor data acquired by another autonomous vehicle.

9. A computing system comprising:

one or more processors; and

one or more tangible, non-transitory, computer readable media that collectively store instructions that when executed by the one or more processors cause the computing system to perform operations, the operations comprising:

obtaining sensor data associated with an environment of a first autonomous vehicle;

determining an intermediate environmental representation of at least a portion of the environment of the first autonomous vehicle based at least in part on the sensor data and a machine-learned model;

generating a compressed intermediate environmental representation by compressing the intermediated environmental representation of at least the portion of the environment of the first autonomous vehicle;

selecting a second autonomous vehicle to which to communicate the compressed intermediate environmental representation from among a plurality of autonomous vehicles based at least in part on an ability of the second autonomous vehicle to decompress the compressed intermediate environmental representation, wherein selecting the second autonomous vehicle comprises:

communicating with the second autonomous vehicle as the second autonomous vehicle enters a communication range of the first autonomous vehicle;

exchanging, with the second autonomous vehicle, data indicating that the second autonomous vehicle is able to decompress the compressed intermediate environmental representation; and

selecting the second autonomous vehicle based on the data indicating that the second autonomous vehicle is able to decompress the compressed intermediate environmental representation; and

communicating the compressed intermediate environmental representation to the second autonomous vehicle.

10. The computing system of claim 9 , wherein the sensor data comprises three-dimensional LIDAR point cloud data.

11. The computing system of claim 9 , wherein determining the intermediate environmental representation comprises:

generating voxelized sensor data based at least in part on the sensor data; and

generating the intermediate environmental representation based at least in part on the voxelized sensor data and the machine-learned model.

12. The computing system of claim 9 , wherein the second autonomous vehicle is configured to decompress the compressed intermediate environmental representation and utilize the intermediate environmental representation for one or more autonomous operations of the second autonomous vehicle.

13. The computing system of claim 9 , wherein the operations further comprise:

obtaining, from another autonomous vehicle, a second intermediate environmental representation of at least the portion of the environment of the first autonomous vehicle.

14. An autonomous vehicle comprising:

one or more sensors;

one or more processors; and

one or more tangible, non-transitory, computer readable media that collectively store instructions that when executed by the one or more processors cause the one or more processors to perform operations, the operations comprising:

obtaining, via the one or more sensors, sensor data associated with an environment of the autonomous vehicle;

determining a first intermediate environmental representation of at least a portion of the environment of the autonomous vehicle based at least in part on the sensor data;

generating a first compressed intermediate environmental representation by compressing the first intermediate environmental representation of at least the portion of the environment of the autonomous vehicle;

determining a recipient to which to communicate the first compressed intermediate environmental representation from among a plurality of potential recipients based at least in part on an ability of the recipient to decompress the compressed intermediate environmental representation, wherein determining the recipient comprises:

exchanging, with the recipient, data indicating that the recipient is able to decompress the compressed intermediate environmental representation; and

selecting the recipient based on the data indicating that the recipient is able to decompress the compressed intermediate environmental representation;

communicating the first compressed intermediate environmental representation to the recipient.

15. The autonomous vehicle of claim 14 , wherein determining the first intermediate environmental representation based at least in part on the sensor data comprises:

generating voxelized sensor data by voxelizing three-dimensional point cloud data of the sensor data;

inputting the voxelized sensor data into a machine-learned model; and

receiving the first intermediate environmental representation as an output of the machine-learned model.

16. The autonomous vehicle of claim 15 , wherein the machine-learned model is configured to apply one or more convolutional layers to the voxelized sensor data.

17. The autonomous vehicle of claim 14 , wherein the operations further comprise:

obtaining a second compressed intermediate environmental representation from another autonomous vehicle;

generating a decompressed intermediate environmental representation by decompressing the second compressed intermediate environmental representation;

determining, using one or more machine-learned models, an updated intermediate environmental representation based at least in part on the decompressed intermediate environmental representation and the first intermediate environmental representation generated by the first autonomous vehicle; and

generating an autonomy output for the autonomous vehicle based at least in part on the updated intermediate environmental representation.

18. The autonomous vehicle of claim 17 , wherein the operations further comprise compensating for a time delay between the first intermediate environmental representation and the decompressed intermediate environmental representation.