The moon has a branch office in Bremen. Inside, a vast crater forms the focal point of an artificial moonscape created to put tomorrow’s astronauts through their paces. This is the space exploration hall at the German Research Centre for Artificial Intelligence (DFKI), and the astronauts in question are robots.
These experimental machines are here to practice independent exploratory missions, roaming craters and their surroundings to analyse sediment layers and trace material from the solar system. The creator of the climbing robot is Professor Frank Kirchner, who heads the Robotics Innovation Centre at DFKI. His creatures are often biologically inspired, such as the four-legged walking robot Charlie, which looks like a monkey, or the six-limbed Mantis, a contraption that looks unnervingly like its insect namesake. At present, Coyote III, a grey-and-orange rover with star-shaped wheels and a flattish silhouette, is carefully navigating the crater.
Intelligent and autonomous robots are indispensable for space exploration because they require no food and no oxygen. And once the mission is done, they don’t need a return journey to Earth. They do, however, have to be able to hold their own up there. The artificial crater in Bremen provides an opportunity to see if they can cut it. The crater was built by a company that usually builds indoor climbing walls, using photographs taken by Apollo astronauts of the Moon’s south pole.
The research group in Bremen has been working intensively on topics such as sensor technology and artificial intelligence, and the results they achieve do not only benefit aerospace application. Kirchner also places great store by the transfer to other fields and is following the development of autonomous vehicles closely.
And there are many commonalities. Both autonomous vehicles and robots on distant moons must perceive and analyse their surroundings and use that information to make intelligent decisions. Of course, on the moon and Mars there are no traffic lights or pedestrians appearing out of nowhere. But Kirchner’s robots have to deal with changing conditions such as sandstorms and tornadoes on Mars or starkly variable light conditions on the moon.
In contrast to autonomous vehicles, however, there are no maps of the terrain for their missions. “At one meter, the resolution of satellite images is still too poor,” explains Kirchner. “As such, the robots must build their own maps of their environment and locate themselves within it.” To cope with that reality, the researchers developed the SLAM algorithms (Self Localization and Mapping), probability-based methods for orientation in unknown terrain.
“It all started with navigation in sewage canals,” recalls Kirchner. “It was a very simple environment, which allowed us to test the new approach there very effectively.” From the mid-1990s, the SLAM algorithms were also used in open terrain and in buildings. The first applications for the self-localization of autonomous vehicles emerged about 15 years ago.
“In robotics, object recognition has gained a great deal in terms of maturity and robustness,” says Kirchner. “The underlying mathematics is the same as in cars today.”
Based on his own research, he knows how complicated it is to steer a car through traffic without human intervention. And as a highly engaged observer of the development, he naturally has a few ideas of his own on the subject.
“Autonomous vehicles should learn during their use phase,” he suggests. “One buys a vehicle with basic experience and it continues to develop itself along with the other vehicles on the road.” It would be a collective learning experience — just as with the collaborative robots that are now gaining a foothold in industrial manufacturing processes: they have to get along with a variety of different people and therefore share their individual experiences with each other.
“Today, with autonomous driving we pay too much attention to the individual algorithms — but we’ve known them for a long time already, in some cases since the 1950s,” says Kirchner. “What’s more important is the organisation of knowledge. The key is to network the individual components of knowledge with each other — for instance through collective learning. The vehicle must be a system that learns throughout its life.”
It will be fascinating to see what ideas from the Bremen-based robotics experts eventually make their way from moon to Earth. Kirchner’s team is already working on the EU project Dreams4Cars, which aims to improve the safety of autonomous vehicles. The control software continually replays real traffic situations in a simulation environment, testing alternative reactions and thereby preparing itself for exceptional circumstances. Fully autonomous driving may still be some way off, but the technology that will enable it is definitely in our orbit.