Anki Learn is a project aimed to add a custom object detection to Cozmo's Anki robots. It is based on project CozmoPedia developed by prof. Dave Touretzky and used in Cognitive Robotics course at Carnegie Mellon University.
For the image classification, I used transfer lerning to over Inception model to train the model with images taken from the Cozmo's onboard caemra.
For image detection, I used slicing technique to divide the image into vertical slices and find the object in them. The processed data would be used by the state machine to direct the robot.
The overall architecture is illustrated here:
