Introduction

Abstract

Since Issac Asimov started to write short stories about robots in the 1940s, the idea of robots as household helpers, companions and soldiers has shaped the popular view of robotics. Science fiction movies depict robots both as friends and enemies of the human race, but in both cases their capabilities far exceed the capabilities of current real robots. Simon (1965), one of the artificial intelligence (AI) research pioneers, claimed that “machines will be capable, within twenty years, of doing any work a man can do.” This over-optimistic promise led to more conservative predictions. Nowadays robots slowly start to penetrate our daily lives in the form of toys and household helpers, like autonomous vacuum cleaners, lawn mowers, and window cleaners. Most other robotic helpers are still confined to research labs and industrial settings. Many tasks of our daily lives can only be performed very slowly by a robot which often has very limited generalization capabilities. Hence, all these systems are still disconnected from the expectation raised by literature and movies as well as from the dreams of AI researchers. Especially in Japan, the need of robotic household companions has been recognized due to the aging population. One of the main challenges remains the need to adapt to changing environments in a co-inhabited household (e.g., furniture being moved, changing lighting conditions) and the need to adapt to individual requirements and expectations of the human owner. Most current products either feature a “one size fits all” approach that often is not optimal (e.g., vacuum robots that are not aware of their surrounding but use an approach for obstacle treatment, that guarantees coverage of the whole floor (BotJunkie, 2012)) or an approach that requires a setup step either in software (e.g. providing a floor map) or in hardware (e.g., by placing beacons). As an alternative one could imagine a self-learning system. In this book, we will not treat navigation problems but rather focus on learning motor skills (Wulf, 2007). We are mainly interested in motor skills that need to take into account the dynamics of the robot and its environment. For these motor skills, a kinematic plan of the movement will not be sufficient to perform the task successfully. A motor skill can often be represented by a motor primitive, i.e., a representation of a single movement that is adapted to varying situations (e.g., a forehand or a backhand in table tennis that is adapted to the ball position and velocity).We focus on learning how to perform a motor primitive optimally and how to generalize it to new situations. The presented tasks correspond to games and sports, which are activities that a user might enjoy with a robot companion, but the presented techniques could also be applied to more mundane household tasks.