The Self-Learning Robot Hand

The Self-Learning Robot Hand

Jun 9, 2017 @ 12:30 |


Researchers at the University of Bielefeld have developed a gripping system with robotic hands, which independently acquaint itself with unknown objects.


The new system works without first knowing the features of objects such as fruit or tools. The Greif-Lern-System was developed in the large project “Famula” of the excellence cluster Cognitive Interaction Technology (CITEC) at the University of Bielefeld. The knowledge from the project could, for example, help future servicerobots to get themselves into new budgets. CITEC is investing around € 1 million for Famula. In a new research_tv contribution by the University of Bielefeld, the project leaders explain the new development.

“Our system learns by experimenting with and exploring itself – just as babies devote themselves to new objects,” says Professor Dr. Helge Knight. The neuroinformatics scientist conducts the project together with the sports scientist and cognitive psychologist Professor Dr. Thomas Schack and the robotics professor Dr. Sven Wachsmuth.

The CITEC scientists work with a robot with two hands, the human hands in form and mobility are modeled. The robot brain for these hands has to learn how everyday objects, such as fruits, dishes, or even plush toys, can be distinguished by their colors and shapes, and what is important when you want to grab them.


Related articles:


Man as a model

A banana can be embraced, a button can be pressed. “The system learns, To recognize such possibilities from features and build a model for dealing with and recognition, “says Ritter.

For this, the interdisciplinary project combines work in the artificial intelligence with research work in further disciplines. The research group of Thomas Schack, for example, investigated the characteristics of test subjects as significant in the field of gripping actions. In a study, the subjects had to compare the similarity of more than 100 objects. “It was surprising that the weight hardly plays a role. We people rely on shape and size when we differentiate things, “says Thomas Schack. In another study, test persons were able to connect their eyes and handle dice that differed in weight, shape and size. Infrared cameras recorded the hand movements. “This is how we learn how people feel an object and which strategies they prefer to use to capture their properties,” says Dirk Koester, a member of Schack’s research group.

“We also learn what mistakes people make when they are blinded.” System moves into the position of its “mentor

In front of a large metal cage with two robotic arms and a table with different test objects, Dr. Robert Haschke, employee of Helge Ritter. He helps the system in the role of a human learning mentor in learning new objects. So he tells the hands what object on the table they are to inspect next. For this purpose, Haschke points to individual objects or linguistic notes, Such as a direction (“rear left”), in which an interesting object can be found for the robot. Two monitors show how the system perceives its environment through color cameras and depth sensors and reacts to the commands of humans.

“The hands must be able to interpret oral language as well as gesture in order to understand with which object they are to deal,” explains Sven Wachsmuth from the CITEC Central Laboratory. “And they must be able to put themselves in the position of the human being, also to inquire whether they have understood correctly.” Wachsmuth and his team are responsible not only for the language competence of the system. They also gave him a face: From one of the screens, Flobi follows the movement of the hands and reacts to the instructions of the scientists. Flobi is a stylized robot head that complements the language and actions of the robot through facial expressions. As part of the Famula system, the virtual version of the robot is currently in use.

Understanding Human Interaction

With Famula, CITEC researchers are conducting basic research that can benefit future household and industrial self-learning robots. “We want to understand how we literally understand our environment through our hands. The robot allows us to check our findings in reality and to uncover gaps in our understanding without a hitch. This is a contribution to the future use of complex, multi-faceted robotic hands, which are still too costly and too complex for use in industry, “says Ritter.

The project name Famula stands for “Deep Familarization and Learning Grounded in Cooperative Manual Action and Language: from Analysis to Implementation” Intensive familiarization and learning with cooperative hand movements and language: from the study to the implementation “. The project has been running since 2014 and is initially limited to October 2017. Eight research groups from the CITEC Excellence Cluster are working on Famula. It is one of four large CITEC projects. The other projects are the robot service apartment, the running robot Hector and the virtual training environment ICSpace. CITEC is funded as part of the Excellence Initiative by the German Research Foundation (DFG) on behalf of the Federal Government and the Länder (EXC 277).


  • Keywords: Self-Learning, Robot Hand, gripping system,CITEC, robot,  robotic arms,artificial intelligence.
  • Source: University of Bielefeld
  • Image: University of Bielefeld

 

Leave a Reply

Your email address will not be published. Required fields are marked *