Training multiple neural networks of a vehicle perception component based on sensor settings

1. A method for controlling a vehicle based on sensor data having variable sensor parameter settings, the method comprising:

receiving sensor data generated by a sensor configured to sense an environment through which a vehicle is moving, wherein:

the sensor comprises a lidar or radar device,

the sensor data includes a point cloud generated by the lidar or radar device, and

the sensor data is generated by the sensor while the sensor is configured with a first sensor parameter setting, wherein the first sensor parameter setting defines a first spatial distribution according to which scan lines of the received point cloud are arranged;

receiving an indicator specifying the first sensor parameter setting;

selecting, by one or more processors and based on the received indicator, one of a plurality of neural networks of a perception component, each of the plurality of neural networks having been trained using training data corresponding to a sensor parameter setting that defines a different spatial distribution of scan lines;

generating, by the one or more processors, signals descriptive of a current state of the environment using the selected neural network and based on the received sensor data;

generating, by the one or more processors, driving decisions based on the signals descriptive of the current state of the environment; and

causing one or more operational subsystems of the vehicle to maneuver the vehicle in accordance with the generated driving decisions.

2. The method of claim 1 , wherein the first spatial distribution of scan lines comprises a uniform distribution, a Gaussian distribution, a multimodal distribution, and/or an arbitrary distribution.

3. The method of claim 1 , wherein:

the first sensor parameter setting further defines a first frame rate at which the received point cloud is captured; and

each of the plurality of neural networks is further trained using training data corresponding to a sensor parameter setting that defines a different frame rate.

4. The method of claim 1 , wherein:

the first sensor parameter setting further defines one or more characteristics of a field of regard of the received point cloud; and

each of the plurality of neural networks is further trained using training data corresponding to a sensor parameter setting that defines different field of regard characteristics.

5. The method of claim 1 , wherein generating the driving decisions based on the signals descriptive of the current state of the environment includes:

generating, by the one or more processors, signals descriptive of one or more future states of the environment based on the signals descriptive of the current state of the environment; and

generating the driving decisions based on both (i) the signals descriptive of the current state of the environment and (ii) the signals descriptive of the one or more future states of the environment.

6. The method of claim 1 , wherein causing one or more operational subsystems of the vehicle to maneuver the vehicle in accordance with the generated driving decisions includes providing data indicative of the generated driving decisions to the one or more operational sub systems.

7. A non-transitory computer-readable medium storing thereon instructions executable by one or more processors to implement a self-driving control architecture of a vehicle, the self-driving control architecture comprising:

a perception component including a plurality of neural networks, each of the plurality of neural networks having been trained using training data corresponding to a sensor parameter setting that defines a different spatial distribution of scan lines, and the perception component being configured to:

receive sensor data generated by a sensor configured to sense an environment through which the vehicle is moving, wherein:

the sensor comprises a lidar or radar device,

the sensor data includes a point cloud generated by the lidar or radar device, and

the sensor data is generated by the sensor while the sensor is configured with a first sensor parameter setting, wherein the first sensor parameter setting defines a first spatial distribution according to which scan lines of the received point cloud are arranged,

receive an indicator specifying the first sensor parameter setting,

select, based on the received indicator, one of the plurality of neural networks, and

generate signals descriptive of a current state of the environment using the selected neural network and based on the received sensor data; and

a motion planner configured to:

generate driving decisions based on the signals descriptive of the current state of the environment, and

cause one or more operational subsystems of the vehicle to maneuver the vehicle in accordance with the generated driving decisions.

8. The non-transitory computer-readable medium of claim 7 , wherein:

the first sensor parameter setting further defines a first frame rate at which the received point cloud is captured; and

each of the plurality of neural networks is further trained using training data corresponding to a sensor parameter setting that defines a different frame rate.

9. The non-transitory computer-readable medium of claim 7 , wherein:

the first sensor parameter setting further defines one or more characteristics of a field of regard of the received point cloud; and

each of the plurality of neural networks is further trained using training data corresponding to a sensor parameter setting that defines different field of regard characteristics.

10. A vehicle comprising:

a sensor configured to generate sensor data by sensing an environment through which the vehicle is moving, wherein:

the sensor comprises a lidar or radar device,

the sensor data includes a point cloud generated by the lidar or radar device, and

the sensor data is generated by the sensor while the sensor is configured with a first sensor parameter setting, wherein the first sensor parameter setting defines a first spatial distribution according to which scan lines of the received point cloud are arranged;

one or more operational subsystems; and

a computing system configured to:

receive the sensor data,

receive an indicator specifying the first sensor parameter setting,

select, based on the received indicator, one of a plurality of neural networks, wherein each of the plurality of neural networks is trained using training data corresponding to a sensor parameter that defines a different spatial distribution of scan lines,

generate signals descriptive of a current state of the environment using the selected neural network and based on the received sensor data,

generate driving decisions based on the signals descriptive of the current state of the environment, and

cause the one or more operational subsystems to maneuver the vehicle in accordance with the generated driving decisions.

11. The vehicle of claim 10 , wherein:

the first sensor parameter setting further defines a first frame rate at which the received point cloud is captured; and

each of the plurality of neural networks is further trained using training data corresponding to a sensor parameter setting that defines a different frame rate.

12. The vehicle of claim 10 , wherein:

the first sensor parameter setting further defines one or more characteristics of a field of regard of the received point cloud; and

each of the plurality of neural networks is further trained using training data corresponding to a sensor parameter setting that defines different field of regard characteristics.

13. The method of claim 1 , wherein selecting the one neural network of the plurality of neural networks comprises selecting a neural network that was trained using the first sensor parameter setting.

14. The non-transitory computer-readable medium of claim 7 , wherein the first spatial distribution of scan lines comprises a uniform distribution, a Gaussian distribution, a multimodal distribution, and/or an arbitrary distribution.

15. The non-transitory computer-readable medium of claim 7 , wherein selecting the one neural network of the plurality of neural networks comprises selecting a neural network that was trained using the first sensor parameter setting.

16. The vehicle of claim 10 , wherein the first spatial distribution of scan lines comprises a uniform distribution, a Gaussian distribution, a multimodal distribution, and/or an arbitrary distribution.

17. The vehicle of claim 10 , wherein selecting the one neural network of the plurality of neural networks comprises selecting a neural network that was trained using the first sensor parameter setting.