Multi-UAV management

A method of identifying and responding to problem conditions for a plurality of aerial vehicles, the method comprising:

• receiving, by one or more processors of one or more computing devices, from an aerial vehicle of the plurality of aerial vehicles, sensor feedback;
• generating, by the one or more processors, using one or more supervised machine learning (ML) models, and based on the sensor feedback, (i) a plurality of likelihood values corresponding to a plurality of possible problem conditions, wherein each respective likelihood value of the plurality of likelihood values indicates a likelihood of the aerial vehicle experiencing a corresponding problem condition of the plurality of possible problem conditions and (ii) a plurality of remaining flight metrics corresponding to the plurality of possible problem conditions, wherein each respective remaining flight metric of the plurality of remaining flight metrics is specific to the aerial vehicle experiencing a corresponding problem condition of the plurality of possible problem conditions, and wherein each of the one or more supervised ML models has been trained using training data generated by the plurality of aerial vehicles having a plurality of different physical configurations that differ from one another by at least one physical component;
• identifying, by the one or more processors, for the aerial vehicle, a particular problem condition of the plurality of possible problem conditions based on comparing (i) the plurality of likelihood values corresponding to the plurality of possible problem conditions to (ii) a plurality of threshold values that is specific to a particular physical configuration of the aerial vehicle, wherein the particular physical configuration of the aerial vehicle is one of the plurality of different physical configurations, wherein the plurality of thresholds are based on a subset of the training data corresponding to an aerial vehicle subset of the plurality of aerial vehicles, and wherein each aerial vehicle of the aerial vehicle subset has the particular physical configuration;
• determining, by the one or more processors, a mitigation response for the particular problem condition based on (i) a mission assigned to the aerial vehicle and (ii) the remaining flight metric generated by the one or more supervised ML models for the particular problem condition;
• and
• sending, by the one or more processors, the mitigation response to the aerial vehicle in order to cause the aerial vehicle to maneuver according to the mitigation response and thereby automatically respond to the particular problem condition.

A non-transitory computer-readable storage medium having stored thereon instructions that, when executed by a computing device, cause the computing device to perform operations comprising:

• receiving, from an aerial vehicle of a plurality of aerial vehicles, sensor feedback;
• generating, using one or more supervised machine learning (ML) models and based on the sensor feedback, (i) a plurality of likelihood values corresponding to a plurality of possible problem conditions, wherein each respective likelihood value of the plurality of likelihood values indicates a likelihood of the aerial vehicle experiencing a corresponding problem condition of the plurality of possible problem conditions and (ii) a plurality of remaining flight metrics corresponding to the plurality of possible problem conditions, wherein each respective remaining flight metric of the plurality of remaining flight metrics is specific to the aerial vehicle experiencing a corresponding problem condition of the plurality of possible problem conditions, and wherein each of the one or more supervised ML models has been trained using training data generated by the plurality of aerial vehicles having a plurality of different physical configurations that differ from one another by at least one physical component;
• identifying, for the aerial vehicle, a particular problem condition of the plurality of possible problem conditions based on comparing (i) the plurality of likelihood values corresponding to the plurality of possible problem conditions to (ii) a plurality of threshold values that are specific to a particular physical configuration of the aerial vehicle, wherein the particular physical configuration of the aerial vehicle is one of the plurality of different physical configurations, wherein the plurality of thresholds are based on a subset of the training data corresponding to an aerial vehicle subset of the plurality of aerial vehicles, and wherein each aerial vehicle of the aerial vehicle subset has the particular physical configuration;
• determining a mitigation response for the particular problem condition based on (i) a mission assigned to the aerial vehicle and (ii) the remaining flight metric generated by the one or more supervised ML models for the particular problem condition;
• and
• sending the mitigation response to the aerial vehicle in order to cause the aerial vehicle to maneuver according to the mitigation response and thereby automatically respond to the particular problem condition.