Adjusting vehicle models based on environmental conditions
Techniques for adjusting vehicle models based on environmental conditions are discussed herein. The techniques may include receiving image data representing a portion of an environment in which a vehicle is operating and inputting the image data into a machine learned model. Additionally, data representing an environmental condition associated with the environment may be received from a sensor of the vehicle to detect changes in the environmental conditions such that one or more actions associated with the machine learned model or an output of the machine learned model may be performed. Some of the techniques may also include running multiple machine learned models or multiple configurations of a machine learned model in parallel and selecting different outputs of the machine learned model(s) to adjust for changes in the environmental conditions. For instance, individual outputs may be selected based on environmental conditions, confidence scores, thresholds, etc.
1. A system comprising:
one or more processors; and
one or more non-transitory computer-readable media storing instructions executable by the one or more processors, wherein the instructions, when executed, cause the system to perform operations comprising:
receiving, from an ambient light sensor of an autonomous vehicle, ambient light data representing an amount of ambient light, at a first time, within an environment in which the autonomous vehicle is operating;
determining, based at least in part on the ambient light data at the first time, that the amount of ambient light within the environment has changed from a second time prior to the first time;
determining, as a similarity, a subset of weights associated with a current configuration of a machine learned model and also associated with a different configuration of the machine learned model;
reconfiguring, based at least in part on the change in the amount of ambient light within the environment and the similarity, the current configuration of the machine learned model to include a subset of weights associated with the different configuration of the machine learned model and not previously associated with the current configuration of the machine learned model, the current configuration of the machine learned model and the different configuration of the machine learned model sharing a same machine learned model architecture associated with multiple available sets of weights;
wherein reconfiguring the current configuration of the machine learned model includes determining a first confidence level associated with data representing a change in the amount of ambient light within the environment;
receiving, from an image sensor associated with the autonomous vehicle, image data representing a portion of the environment;
inputting the image data into the machine learned model; and
performing an action associated with at least one of the machine learned model or an output of the machine learned model.
2. The system of claim 1 , wherein the amount of ambient light at the first time is associated with an operational amount of ambient light associated with the image sensor of the autonomous vehicle.
3. The system of claim 1 , the operations further comprising:
determining that the amount of ambient light at the first time meets a threshold amount of ambient light.
4. The system of claim 3 , wherein the current configuration of the machine learned model differs from the different configuration of the machine learned model in at least one of:
a threshold associated with the machine learned model; or
a second confidence level associated with an aspect of the output of the machine learned model.
5. The system of claim 1 , the operations further comprising adjusting, based at least in part on the determining that the amount of ambient light within the environment has changed from the second time prior to the first time, a lens of the image sensor of the autonomous vehicle.
6. A method comprising:
receiving, from a photosensor of a vehicle, data representing an environmental condition associated with an environment;
determining, based at least in part on the data representing the environmental condition associated with the environment, a change in the environmental condition;
determining, as a similarity, a subset of weights associated with a current configuration of a machine learned model that includes the subset of weights and also associated with a different configuration of the machine learned model that includes the subset of weights;
adopting, based at least in part on the change in the environmental condition and the similarity, the different configuration of the machine learned model that includes the subset of weights, wherein adopting the different configuration of the machine learned model that includes the subset of weights comprises determining a confidence level associated with the data representing the change in the environmental condition;
receiving image data representing a portion of the environment in which the vehicle is operating;
inputting the image data into the machine learned model; and
performing an action associated with the vehicle based on output received from the machine learned model.
7. The method of claim 6 , wherein the environmental condition comprises ambient light in the environment.
8. The method of claim 7 , further comprising:
determining that the data representing the environmental condition associated with the environment indicates less than a threshold light level, wherein the threshold light level is based at least in part on an operational parameter associated with an image sensor capturing the data.
9. The method of claim 6 , wherein:
the current configuration of the machine learned model that includes the subset of weights comprises a first configuration of the machine learned model associated with the environmental condition; and
the different configuration of the machine learned model that includes the subset of weights comprises a second configuration of the machine learned model associated with a differing environmental condition.
10. The method of claim 9 , wherein performing the action associated with the vehicle comprises changing a threshold associated with the machine learned model.
11. The method of claim 9 , wherein the current configuration of the machine learned model that includes the subset of weights and the different configuration of the machine learned model that includes the subset of weights further differ based on a threshold associated with the machine learned model.
12. The method of claim 6 , wherein performing the action associated with the vehicle comprises performing a receding horizon technique to generate a plurality of paths for the vehicle to traverse.
13. The method of claim 6 , further comprising:
determining an ambient lighting state associated with the image data; and
adopting the different configuration of the machine learned model that includes the subset of weights based at least in part on the ambient lighting state associated with the image data.
14. One or more non-transitory computer-readable media storing instructions that, when executed, cause one or more processors to perform operations comprising:
receiving, from a photosensor of a vehicle, data representing an environmental condition associated with an environment;
determining, based at least in part on the data representing the environmental condition associated with the environment, a change in the environmental condition;
determining, as a similarity, a subset of weights associated with a current configuration of a machine learned model and also associated with a different configuration of the machine learned model, wherein the current configuration of the machine learned model and the different configuration of the machine learned model share a same machine learned model architecture;
adopting, based at least in part on the change in the environmental condition and the similarity, the different configuration of the machine learned model;
receiving image data representing a portion of an environment in which a vehicle is operating;
inputting the image data into the machine learned model; and
performing an action associated with the vehicle based on output received from the machine learned model.
15. The one or more non-transitory computer-readable media of claim 14 , wherein the environmental condition comprises ambient light in the environment.
16. The one or more non-transitory computer-readable media of claim 15 , further comprising:
determining that the data representing the environmental condition associated with the environment indicates less than a threshold light level, wherein the threshold light level is based at least in part on an operational parameter associated with an image sensor capturing the data.
17. The one or more non-transitory computer-readable media of claim 14 , wherein:
the current configuration of the machine learned model comprises a first configuration of the machine learned model associated with the environmental condition;
the different configuration of the machine learned model comprises a second configuration of the machine learned model associated with a differing environmental condition; and
the adopting, based at least in part on the change in the environmental condition and the similarity, includes reconfiguring the first configuration of the machine learned model to include a subset of weights associated with the different configuration of the machine learned model and not previously associated with the current configuration of the machine learned model.
18. The one or more non-transitory computer-readable media of claim 17 , wherein the current configuration of the machine learned model and the different configuration of the machine learned model differ based on a threshold associated with the machine learned model.
19. The one or more non-transitory computer-readable media of claim 14 , wherein performing the action associated with the vehicle comprises performing a receding horizon technique to generate a plurality of paths for the vehicle to traverse.
20. The one or more non-transitory computer-readable media of claim 14 , the operations further comprising:
determining an ambient lighting state associated with the image data; and
adopting the different configuration of the machine learned model based at least in part on the ambient lighting state associated with the image data.