Traffic light identification and/or classification for use in controlling an autonomous vehicle

A method implemented by one or more processors of an autonomous vehicle, the method comprising:

• determining, based on a pose instance of the autonomous vehicle and a stored mapping of an environment:
• that a region of the environment includes a traffic light, and
• a configuration that is assigned to the traffic light, the configuration being one of multiple disparate configurations in a taxonomy of configurations, and the configuration defining an orientation of the traffic light and a bulb pattern of the traffic light;
• responsive to determining that the region includes the traffic light:
• processing an image, that captures the region of the environment that includes the traffic light, to generate predicted output associated with multiple candidate states of the traffic light, wherein the predicted output associated with the multiple candidate states of the given traffic light includes a corresponding numerical measure for each of the multiple candidate states for the configuration that is assigned to the given traffic light, and wherein processing the image to generate the predicted output associated with the multiple candidate states of the traffic light comprises:
• selecting, based on the configuration, a machine learning classifier from among a plurality of candidate machine learning classifiers for each of the multiple disparate configurations in the taxonomy of configurations;
• and
• processing, using the machine learning classifier, the image to generate the predicted output associated with the multiple candidate states of the given traffic light;
• selecting one of the candidate states as a current state of the traffic light, wherein the selecting the one of the candidate states as the current state of the traffic light is based on the predicted output;
• and
• controlling the autonomous vehicle based on the selected current state of the traffic light.

A traffic light system of an autonomous vehicle, the traffic light system comprising:

• at least one processor;
• and
• at least one memory comprising instructions that, when executed, cause the at least one processor to:
• capture an original image of an environment of the autonomous vehicle, the original image including a plurality of traffic lights;
• crop the original image to generate a cropped image, the cropped image including a region of the environment that includes a given traffic light of the plurality of traffic lights;
• determine a configuration that is assigned to the given traffic light, wherein the configuration that is assigned to the given traffic light is one of multiple disparate configurations in a taxonomy of configurations, and wherein the configuration that is assigned to the given traffic light defines an orientation of the given traffic light and a bulb pattern of the given traffic light;
• process the cropped image, that captures the region of the environment that includes the given traffic light, to generate predicted output associated with multiple candidate states of the given traffic light, wherein the predicted output associated with the multiple candidate states of the given traffic light includes a corresponding numerical measure for each of the multiple candidate states for the configuration that is assigned to the given traffic light, and wherein the processing the cropped image to generate the predicted output associated with the multiple candidate states of the given traffic light comprises instructions to:
• select, based on the configuration that is assigned to the given traffic light, a machine learning classifier from among a plurality of candidate machine learning classifiers for each of the multiple disparate configurations in the taxonomy of configurations;
• and
• process, using the machine learning classifier, the cropped image to generate the predicted output associated with the multiple candidate states of the given traffic light;
• select one of the candidate states as a current state of the given traffic light, wherein the selecting the one of the candidate states as the current state of the given traffic light is based on the predicted output;
• and
• control the autonomous vehicle based on the selected current state of the traffic light.