Prof. Lilienthal, you have been a professor at TUM since July 2022. What topics do you cover in your new Chair?
I will deal primarily with three topics, the sense of smell of robots in the context of environment and sustainability; with mobile, moving robots and how they can safely and efficiently navigate specifically in industrial environments; as well as the use of Artificial Intelligence (AI), specifically how AI can build on existing domain knowledge instead of having to learn it over and over again.
What are robots that can smell good for?
These robots react to gases and are able to detect methane, carbon monoxide, and other toxic gases, for example. At a mine in Kiruna in Sweden, for example, underground diesel-powered machines were converted to electric operation. This requires very heavy powerful batteries. To detect early when they are running hot or even catching fire, our robots have sensors that sense toxic gases. In addition, we want to find geothermal sources with the help of "Gasbots". For this, we draw gas maps from the surface. From the distribution of the detected gases, it is possible to find out where suitable sites for geothermal wells are located. In the long term, our goal is to automate the measuring and mapping and to hand over the interpretation of the data to an artificial intelligence. Teaching a robot to smell is not easy. The robot must be able to be precisely localized, we must interpret the data from the sensors, use machine learning, transmit signals reliably, and, last but not least, bring the physics of gas propagation into the interpretation. With the sense of smell for robots, the Munich Institute of Robotics and Machine Intelligence (MIRMI) will in the future cover another human sense in addition to sight and touch and will increasingly become the "Center of the Senses".
In addition, at TUM you will be involved with mobile, moving robots ...
For this, perception is the starting point, i.e. the precise perception of the environment so that a robot can ultimately navigate safely and efficiently. An important field of application is industrial environments, very dynamic environments, especially if the place of use is, for example, a warehouse that has grown historically and was not thought for automation from the beginning. At Örebro University, we have deliberately chosen a "semi-controlled environment" for the research, where the vehicles travel more slowly than, say, cars, and changes can be introduced that can make subtasks easier for the robot. To guarantee highly reliable localization, patterns can be placed on the ground, for example. If we had gone straight into traffic, we would have had too many challenges at once. Now we are using our findings for use in factories - and of course in the MIRMI lighthouse project KI.FABRIK.
What role does artificial intelligence play in your research?
In my research, AI is always used where there are dependencies that are difficult to model without AI. Here, with my group and in collaboration with DLR, I am trying to develop particularly data-efficient and reliable approaches in which AI builds on existing domain knowledge. For example, a smelling robot should "know" the partial differential equations from physics that describe the dispersion of gases. This can then be used, as in the example of the search for good geothermal sources, to find out quite reliably in an automated way where to measure next. Another example is eye movement tracking. For example, if a robot wants to walk through a crowd of people without constantly colliding with someone, it helps that it analyzes the people's line of sight. On this basis, he can decide which path to take. There are already startups developing technology to do this. AI plays an important role in this, both in tracking gaze and in interpreting a gaze sequence.
You are one of the world's leading professors of smelling robots and have now moved from Örebro University to TUM. What particularly attracts you to TUM?
In Örebro, about 80 to 90 experts were working on AI and robotics. This was a strong unit, however, there were "only" environmental chemists besides us who were internationally competitive. For an interdisciplinary exchange with physicists or mathematicians, for example, it was always necessary to look for cooperation partners from other universities. At TUM, I look forward to the interdepartmental exchange within the organization. After all, that is one of the hallmarks of the integrated institute MIRMI, to promote this interdisciplinary exchange. My big goal is to establish a kind of "Robot Olfaction CERN" in Munich, a high-tech laboratory for olfaction in robotics as a focal point for scientific teams from all over the world working on this topic.
Information