Tobias Schöls

Wednesday, May 16, 2018, 14:00

Room 01-012, Georges-Köhler-Allee 102, Freiburg 79110, Germany

Robots play an important role in our economy already, where they manufacture our cars or handle dangerous materials. In recent years an increasing number of mobile robots have been added for example to transport goods, such as your next Amazon order. The number of mobile applications is growing rapidly and has already reached the private sector, where they are mowing lawns and vacuuming homes while people are at work. Many applications require that humans and robots move in a shared space and interact and it is important that robots react to their dynamic environment such that collisions and harm are avoided.

The method developed during this master thesis provides responsive behavior and collision avoidance in dynamic environments. 
An optimization-based approach is used to avoid obstacles and to find the shortest path to the goal. Starting from a grid representation of the robot's surroundings (occupancy grid map) we formulate a non-linear programming (NLP) problem. To compute controls that pilot the robot, the problem is repetitively solved in an non-linear model predictive control (NMPC) setting using multiple shooting and the interior point solver Ipopt.
After tests and evaluation in simulation, the method was applied to a real-world robot. It maneuvered the robot through a group of moving people, preventing all collisions.