TUM: Robot moves safely through crowds of people
Robots should not collide with people when moving around in crowds. To do this, the robot has to learn how humans do it.
Sepehr Samavi and Professor Angela Schoellig together with robot "Jack".
(Image: Astrid Eckert / TUM)
A research team at the Technical University of Munich (TUM) has developed algorithms for a mobile robot that give it the ability to maneuver through crowds of people as safely as a human without colliding with other people. The scientists believe that similar algorithms could also be used for humanoid robots and autonomous driving.
The four-wheeled robot “Jack” is just one example of how robots could reach their destination in the future without colliding with people or cars. Essentially, it's all about the technology and, above all, the algorithms that enable the robot to do this. Jack uses lidar to detect his surroundings, as can be seen in the study “SICNav: Safe and Interactive Crowd Navigation Using Model Predictive Control and Bilevel Optimization”, published in IEEE Transactions on Robotics. The robot uses the reflections from the lidar laser to build a 360-degree map from its perspective. In doing so, it takes into account fixed obstacles, but also moving objects and, above all, people. Sensors in the wheels determine the robot's speed and the distance it has covered so far.
Predicting human movements
The robot uses this information to determine the routes that people are likely to take within the next two seconds. At the same time, it calculates whether and how it might have to change its route to its destination to avoid colliding with them. This happens about ten times a second. Jack learns by adopting the movement behavior of humans.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
Accordingly, the researchers have developed a mathematical model and created a planning algorithm derived from human movements. This means, for example, that the robot does not stop immediately if a collision with a human is imminent. Instead, it reacts similarly to a human and changes its route slightly to prevent a collision. In doing so, the robot considers that the human could also change its route slightly. The aim is for the robot not to stand still, but rather to reschedule its route to a greater extent in an emergency.
Incorporating human behavior patterns
The TUM researchers have included data sets with behavioral patterns of people in crowds. Jack thus copies the behavior of people so that his way of moving around becomes more and more human-like. “Jack knows his destination, observes people and sees where they are going in order to then constantly optimize his paths,” summarizes Professor Angela Schoellig, “almost like a human”. The robot is thus not only able to react to a situation and predict the movements of people, but can also interactively influence people through its behavior and thus avoid collisions.
In principle, the algorithms can be transferred to other robots – including humanoid robots that move on legs instead of wheels. One practical application is in delivery robots that have to share the footpath with humans. The use in electrically powered wheelchairs is also an application scenario.
Videos by heise
However, the researchers are thinking even further ahead and see their algorithms being used in autonomous vehicles. They could close a gap there, as the algorithms can consider the reactions of drivers.
(olb)
