Jens Kober Brings People and Machines Together

Whether helping out around the house, bringing meals to patients, assisting with surgeries, harvesting tomatoes in the greenhouse, lending a hand on a construction site, or stocking shelves at the supermarket, "Cognitive robots need to be capable of more than just performing the same task 24 hours a day on the factory floor," explains Kober, who heads the new Department of Learning and Interactive Robots at the Institute for Artificial Intelligence at the University of Stuttgart.
"They must be aware of their surroundings, react to changing conditions, act independently when needed, adapt to new situations, and work alongside their human counterparts to perform a range of tasks."

Humans are training robots
Cognitive robots are systems based on the complex interaction of various components ranging from sensor technology and speech recognition to control systems and human-machine interaction. In this wide-ranging field, Kober researches and develops AI-based methods, tools, and algorithms that make it possible to further train cognitive robots in their respective areas of application beyond basic programming.
Whether via a keyboard, touchscreen, or even physical contact, he gathers the data needed for this task directly from interactions between people and robots. "We directly involve people in the learning process," says Kober. "They guide the robot in much the same way that coaches guide athletes or teachers guide students. This is extremely important for efficiently getting cognitive systems up and running."

User-friendly programming
The challenge here is not merely to store new information and retrieve it immediately but rather to generalize it and apply it to similar situations. After all, every household, every product in the supermarket, every patient, and every small production run in industrial manufacturing is different.
"The various applications all have one thing in common," says Kober. "There are countless possible variations of each task.” He considers it unrealistic to program robots in advance for every possible scenario in their intended operating environments. That is precisely why he wants to enable the future users and 'colleagues' of these intelligent machines to teach and train their robots themselves. Because they understand the task and its requirements better than anyone else."