Devices, systems, and methods for device-error analysis using computer vision and machine learning

A method comprising:

• obtaining a video of a device, wherein the device includes one or more light emitters that are visible in the video;
• generating a series of time-indexed frames based on the video;
• executing a first machine-learning model using the series of time-indexed frames as inputs, wherein the first machine-learning model outputs a time-indexed series of light-emitting states that indicate respective light-emitting states of the one or more light emitters at respective times in the time-indexed series;
• and
• executing a second machine-learning model using the time-indexed series of light-emitting states as inputs, wherein the second machine-learning model outputs a status indicator of the device.

One or more non-transitory computer-readable storage media storing computer-executable instructions that, when executed by one or more computing devices, cause the one or more computing devices to perform operations comprising:

• obtaining training data, wherein the training data includes time-indexed frames of at least one video of a device, wherein the device includes one or more light emitters, wherein the one or more light emitters are visible in the time-indexed frames, and wherein the training data indicates a respective light-emitting state of each of the one or more light emitters in each of the time-indexed frames;
• training a first machine-learning model based on the training data, wherein the first machine-learning model accepts time-indexed frames as inputs and outputs a time-indexed series of light-emitting states that indicates respective light-emitting states of the one or more light emitters at respective times;
• and
• training a second machine-learning model based on a plurality of time-indexed series of light-emitting states and on data that maps operating statuses of the device to respective time-indexed series of light-emitting states, wherein the operating statuses include one or more error statuses, and wherein the second machine-learning model accepts a time-indexed series of light-emitting states as an input and outputs an operating status of the device.