Flexible visually-driven object classification using the baxter robot

One of the main applications for the robotics industry is the classification and manipulation of manufactured objects to increase productivity. Classical open loop robotic manipulation does not allow for changes in the environment without prior re-programming. To make the system more flexible, sensors are used to reduce the error and improve the efficiency for repeatable tasks. An important improvement consists in using visual feedback to avoid mechanical errors and chaining according to real-time circumstances. This work presents the classification of a group of objects based on their color and shape. The process includes image processing, inverse kinematics, and an automation algorithm which allows the task to be defined by the user or by a specific goal. This approach is validated using the Baxter robot and its internal cameras.