Zurich The US robot manufacturer Boston Dynamics is also known to an audience of millions outside of the engineering scene. The four-legged robotic dog “Spot” is responsible for this: Videos in which the robots dance synchronously went viral on social networks. Spot is considered a prime example of what robots can do today.
CEO Péter Fankhauser founded Anybotics together with engineering colleagues from ETH Zurich in 2016 – the same year that Boston Dynamics presented the first version of its robotic dog. Both “Spot” and “Anymal” are now used for inspections in industrial plants.
“Today we can proudly say: We are on an equal footing,” says ETH professor and co-founder Roland Siegwart. A form of artificial intelligence (AI), or to be more precise: deep learning algorithms, was decisive in enabling an initially small team of engineers and software developers to catch up with Boston Dynamics in development by several years.
AI is the basic technology for inspection solutions to build something that can surpass human sensory organs, Anybotics CEO Fankhauser confirms: “Without artificial intelligence, automation cannot progress.”
Complex interaction of motors and sensors
The further development of autonomous robots is representative of the trend that AI applications act more human-like. The technology achieved its breakthrough by solving problems that humans can hardly cope with, such as processing large amounts of data. “Today, a computer can play chess much better than a human because it is able to plan many more moves in advance,” says Siegwart.
But research is increasingly shifting towards training processes that living beings master intuitively. “When a foal is born, it can already walk,” Siegwart gives an example.
>> Read here: The search for the artificial brain: The hardware for AI could come from Europe
For a robot, the learning process is much more complex, an interaction of motors, sensors, cameras and the radar-like lidar technology. Only the exponential growth in computing power, combined with advances in deep learning algorithms, brought the breakthrough here.
It was a big challenge for Anybotics to teach the robot dog how to walk on uneven ground or climb stairs. The company therefore uses learning algorithms designed as artificial neural networks.
The inspection robot can also reach higher areas of the factory premises.
(Photo: Anybotics)
These calculation models, inspired by the nerve cell connections in the human brain, enable the robot to virtually train millions of complex movement sequences. As a result, Anybotics saved thousands of hours of live practice. “The use of deep learning algorithms in robotics is not yet standard,” says ETH scientist Siegwart. But the potential is huge.
What is already possible today is shown not least by a number of start-ups that have emerged from Professor Siegwart’s ETH Laboratory for Autonomous Systems.
- For example, Sevensense Robotics developed “eyes and brains” for autonomous robots. This enables a cleaning robot, for example, to navigate through a well-stocked department store. Gianluca Cesari, co-founder of Sevensense, says: “Our software helps robots to find out where they are in a room, what is happening around them and how to best find their way from a starting point to a defined destination.”
- The security robot from Ascento independently patrols extensive industrial sites, such as those belonging to the Swiss Federal Railways. He can identify people and check whether they are allowed to stay on the premises.
- Voliro, in turn, is developing a drone that flies through industrial plants on defined routes and can take measurements at great heights.
Christian Noske also observes that the technological advances in AI, especially in deep learning algorithms, are fueling the development of a completely new generation of robots. He is a partner at the venture capital investor NGP Capital, analyzes and invests in robotic solutions for industry. He says: “For a long time, industrial robots were only an issue for large corporations.”
This text is part of the large Handelsblatt special on artificial intelligence. Are you interested in this topic? All texts that have already appeared as part of our theme week
You will find here.
Car manufacturers, for example, have automated their production lines with robots to such an extent that humans do not have to intervene in many process steps, and are not allowed to do so at all. But that’s only worth it on a grand scale. This is where the fundamental difference to the new generation of autonomous robots lies: “They are designed to work in environments in which people also work.” The machines could thus be used in a broader range of companies. “This appeals to customers who are buying a robot for the first time,” says Noske.
Easy to control like an iPhone
CEO Fankhauser is convinced: “In the future, autonomous robots will not only be able to perceive and measure their surroundings, but also take on tasks independently.” One example is taking chemical samples autonomously and analyzing them. He estimates that Anybotics will need about a year to bring such capabilities to market.
Cesari expects from Sevensense: “In five to ten years it will be possible to automate almost every work step in the material handling industry.” He sees the largest market in applications relating to flexible production as well as the storage and distribution of goods. “In the future, a logistics warehouse will need far fewer human workers, and productivity will increase dramatically,” he is convinced.
At the same time, advances in complex language models like ChatGPT are making it easier for humans to interact with robots, says investor Noske. “The development has progressed so much that people will soon be able to communicate with a robot as they would with a colleague.” Sevensense believes in this and works on it. Cesari adds: “Our goal is for a robot to be as easy to use as an iPhone.”
But scientist Siegwart also knows the limits of what machines can achieve in the foreseeable future. “There are many robots that can assemble a car – but none that can repair a car.”
More: Anybotics raises $50 million in venture capital.