Automated object detection in a dusty environment

A system for object identification for an autonomous vehicle, the system comprising:

• a storage subsystem to store a plurality of dust classifiers, each dust classifier trained to distinguish between dust and not-dust, wherein each dust classifier is trained using sensor data collected in a different one of a plurality of environmental categories;
• a set of sensors to collect sensor data from an environment around an autonomous vehicle;
• and
• a processor coupled to the set of sensors and the storage subsystem, the processor configured to:
• receive sensor data from the set of sensors;
• determine a current environmental category for the autonomous vehicle;
• select a dust classifier from the plurality of dust classifiers based on the current environmental category;
• apply the selected dust classifier to the sensor data to determine whether a dusty condition is present;
• apply an object classifier to the sensor data to identify an object in an environment of the autonomous vehicle;
• perform a consistency check on an output of the selected dust classifier and an output of the object classifier;
• and
• make a final object identification based on a result of the consistency check and the output of the object classifier.

A computer-readable storage medium having stored therein program instructions that, when executed by a processor, cause the processor to perform a method comprising:

• receiving sensor data from one or more environmental sensors of an autonomous vehicle;
• determining a current environmental category for the autonomous vehicle;
• selecting a dust classifier from a library of dust classifiers based on the current environmental category, wherein the library of dust classifiers includes a plurality of dust classifiers, each dust classifier trained to distinguish between dust and not-dust, wherein each dust classifier is trained using sensor data collected in a different one of a plurality of environmental categories;
• applying the selected one of the plurality of dust classifiers to the sensor data to determine whether dust is present;
• and
• controlling an action of the autonomous vehicle based at least in part on whether dust is present.