You might say that the LearningGripper from Festo behaves just like a baby. But that’s a good thing. Newborn children gradually—and by trial and error—learn how to hold, move, and turn an object, such as a parent’s finger or a toy, with their hands. They want to look at it from all sides and form a three-dimensional impression of it. The gripper learns to perform motion sequences in much the same way.
Inspired by the human hand, the LearningGripper has four fingers. With help from machine learning software, this gripper can master a complex action like picking up and orienting an article. The basic positions of the fingers and the feedback function from the environment need to be defined in advance; the gripper learns all other motion sequences by trial and error.
The LearningGripper’s task, as illustrated, was to turn the ball until the logo is at the top. At the beginning the gripper moved the ball randomly. A position sensor in the ball provided feedback on how far the logo was from the gripper’s “palm.” The LearningGripper received a reward based on a points system; points are processed in the machine learning software. Over time, the software developed a movement strategy and the gripper learned what action to take at a particular point. It changes its motions to receive as much positive feedback as possible and finally finds a reliable solution to its task. If the strategy of one gripper is transferred to another, the second gripper uses that as a knowledge base to learn its own strategy more efficiently.
The LearningGripper demonstrates how systems in the future will be able to solve intricate tasks autonomously without complex programming. Self-learning systems such as the LearningGripper could be installed on a production line and then allowed to optimize their behavior independently.
Submitted by Frank Langro, Festo Corp., Director – Marketing and Product Management.
Filed Under: AI • machine learning, Pneumatic equipment + components, ENGINEERING SOFTWARE, Robotics • robotic grippers • end effectors