Apprenticeship Learning for Cyber-Physical System Intelligence

摘要

Apprenticeship learning [1] is the process of utilizing a reward function and a dynamics model to allow a robot to learn from an expert performing a task. A difficulty for apprenticeship learning is defining what doing a task "well" is. For example, what is considered "driving well" to one person is totally different to another. To one person, it may be traveling between two points without being involved in an accident. Another person may see it as getting from point A to point B inContents 5.1 Introduction ..................................................................................................................... 63 5.2 Apprenticeship Learning via Inverse Reinforcement Learning ..........................................65 5.3 Exploration of Apprenticeship Learning in Reinforcement Learning ............................... 69 5.4 Gridworld ......................................................................................................................... 70 5.5 Car Driving and Parking Application .............................................................................. 71 5.6 Helicopter Flight with Parameterization ...........................................................................74 5.7 Conclusions ...................................................................................................................... 75 References ..................................................................................................................................7650