Digital Talk

Abstract – Deep Reinforcement Learning

Deep Reinforcement Learning (DRL) has gained considerable momentum in recent years with the advent of new algorithms that allow to learn by trial and error in comparably large and continuous observation and action spaces. Examples range from learning to play Go (AlphaGo) over learning to play Atari games from looking at the Video input to chatGPT that uses DRL to improve its text generation capabilities.
This talk will show how DRL algorithms have pushed the boundaries of what can be learned considerably. The application domain are our simulated humanoid soccer playing robots. Using DRL for behavior learning, the kick distances and run speeds could be more than doubled. Other than pure results, I will also share some lessons learned while exploring this exciting new field.