"We first built on existing mechatronics technology: a small crawling robot to inspect the low-pressure network of gas pipelines. With this Snakebot, called Pirate, we started working in Smart Tooling, together with ExRobotics and our PhD student Nicolò Botteghi. Thanks to Smart Tooling we were able to finance it. Nicolò was especially committed to the intelligence of the robot. The morphology was in order, so we as a department did not pursue this further. However, it was the intention that the robot would work autonomously, not with a remote. This means, you have to work on artificial intelligence." THE REAL WORLD "In practice, AI has to deal with everchanging, unpredictable circumstances. In a virtual environment, or a physical environment that is fairly stable, for example a room, AI is quite simple. This also applies to a perfectly clean pipe. But the situation in the processing industry is different. Sometimes there is oil in the pipeline, or other pollution. You don't know where or when. This means the robot has to learn how to turn his 'body' with slipping wheels". "So the expertise has originally grown from the mechatronics side, the design of the robot itself. Good engineering produces a robust robot that crawls through a small pipe. Thanks to Smart Tooling, we were able to pay a lot of attention to the fundamental problem of autonomy, i.e. intelligence. How can you create intelligence in a robot? How should it learn? We performed many tests in a simulation environment. We were not able to test on a real robot, in practice. But we were able to generate knowledge for the next steps in automating crawling robots in pipelines". "The basic concept for 'learning' based on unknown factors is now there. So the next step is to work with a company with a real robot again. Hopefully in a next project." COMPLEX "In the meantime, AI is also being further developed. Reinforced and deep learning, for example, are growing in the medical world and this also broadens our horizon. But beware ... 99% of people say: AI is the solution to all problems. That is not just the case. With a few lines of software you get nowhere. What we want is much more complicated than playing chess or GO with a computer. Getting a small robot to come up with something on its own is much more complex. In a virtual world you know everything, so there are algorithms. That is a model, not reality. Moving and reacting in and to a physical world is much more complex. What is going to happen is not clear, so you have to measure and decide in the moment, learn from your own mistakes and the environment you don't know. That is where the challenge lies for us. FUTURE "This interaction between body and mind is becoming more and more interesting because more and more computational power is becoming available for intelligence in robotics. That is an important element. 48
51 Online Touch Home