PSort-Bot
PSort-Bot automates the pick-inspect-place tasks guided by computer vision, which are employed in our nation’s postharvest fruit and vegetable sorting lines. It seamlessly integrates a computer vision module, which quickly detects common blemishes on sweet yellow onions with very high accuracy. Toward realizing this, InversAI has assembled one of the largest image data sets of blemished and unblemished Vidalia onions under various conditions, which is used for training the model.
PSort-Bot integrates with a standard depth camera and a collaborative robot manipulator to engage in sorting unblemished onions from blemished ones on your existing conveyance lines. It installs with minimum fuss and easily scales up to the needed processing throughput by deploying multiple PSort-Bots, which work together as a team offering superlinear speedup.
PSort-AR
PSort-AR augments the produce on processing lines with AI-based predictions. Developed in response to feedback from a robot integrator and a farm owner, PSort-AR lets our AI blemish detection model self-learn "in the wild" with data from actual sortation, and shows the model’s predictions to the line workers. It lets others know which onions the robot will pick, and that they should pick if the robot is unable. A secondary benefit of such transparency is that it promotes automation acceptability. PSort-AR represents an intermediate stage in the staged integration of PSort-Bot into produce processing lines.
Once the AI blemish detection model is robust and more accurate, PSort-AR can be used to train new line workers, which is useful in packing sheds with high personnel turnover. It will save supervisor time otherwise spent on frequent trainings, and reduce human subjectivity.
