Research area: Autonomous Robots: Robotic Soccer and beyond

Making robots act autonomously in a complex environment bears many challenges: uncertainties in sensor and actors, many degrees of freedom, constraints to be considered, etc.

We therefore consider learning robots to be a highly interesting application domain for machine learning techniques. In particular we are interested in reinforcement learning controllers: the designer only specifies the goal and the actions; the robot then learns the appropriate behavior by (clever) trial and error.

Some concrete research projects within this area are:

• Robotic Soccer: The Brainstormers (since 1998)
• Reinforcement Controllers in Technical Applications (since 1994)
• Learning Algorithms for Cooperative Multi-Agent Systems (DFG; 2004-2006)
• RL in real, distributed, and cooperative environments (DFG SPP 1125; 2001- 2006)
• CLSquare: Closed Loop Simulation System for learning controller (since 2004)
• Reinforcement Learning in Realtime (DFG SFB 531; 2003-2004)
• Learning on Varying Time Scales (DFG; 1998-2000)
• Fynesse: Fuzzy Neuro Systems for Control (DFG; 1996-2000)
• OpenTribot: Design of a standard plattform (hard- and software) for the RoboCup Midsize League in an Open-Source project (DFG RI 923/6-1; 2009-2012)

See also our publications page for further information.