How automation, machine learning, and DNA barcoding can accelerate species discovery in “dark taxa”: Robotics and AI

Robotics and artificial intelligence are two methods that are suitable for improving processes that are normally done manually. Therefore, these techniques also can be used when examining specimen-rich invertebrate samples, where traditional sorting methods are to slow and require expert knowledge. For that reason, we developed the DiversityScanner: a classification, sorting, and measurement robot for invertebrates. The 500 x 500 x 500 mm robot has three linear axes that enable a camera unit and an automated pipette to be moved over a square Petri dish, containing up to 150 specimens. After starting the DiversityScanner the image taken by an overview camera mounted directly above the Petri dish is utilized to calculate the position of the insects. Then the camera unit is moved over one specimen to capture high resolution detailed images. Convolutional neuronal networks (CNNs) are then used to classify the specimen into 14 different insect taxa (mostly families) and the specimen length and volume are estimated. In a final step, the specimen is moved into a microplate using an automated pipette. In this talk we show how the DiversityScanner uses automation and artificial intelligence to take advantage of previously nearly untapped resources in the study of specimen-rich invertebrate samples.