| Abstract |
Recent developments within the field of deep learning have resulted in novel algorithms that demonstrated the high potential reinforcement learning (RL) has as a learning framework. Most applications of RL rely on simulation environments (i.e. Atari games, Go, etc..) where large amounts of data and interactions are available. To bridge the gap towards the application of RL techniques for solving real-world problems, we developed a simulation environment for daily human activities. This simulator is powered by generative adversarial networks (GANs) that synthesize low-level sensor data and is presented as a reliable testbed for novel RL algorithms. We aim to develop data efficient end-to-end RL algorithms that can be deployed in real world scenario’s such as the health domain to achieve high levels of personalization towards end users. |