Planning with Expectation Models

TL;DR: The problem of credit assignment in reinforcement learning is addressed and fundamental questions regarding the way in which an agent can best use additional computation to propagate new information are explored, by planning with internal models of the world to improve its predictions.

TL;DR: The proposed two-stage control method includes a decision-making module computing the high-level lane change action and a lateral control module outputting the low-level steering angle which can improve the data efficiency and speed up the training process.

TL;DR: In this paper, a low-dimensional encoding of the environment is learned with a combination of model-based and model-free objectives, and intrinsic rewards that are based on the distance of nearest neighbors in the low dimensional representational space are used to gauge novelty.

TL;DR: In this paper , the authors propose subtasks that use the original reward plus a bonus based on a feature of the state at the time the option stops, and show that options and option models obtained from such reward-respecting subtasks are much more likely to be useful in planning and can be learned online and off-policy using existing learning algorithms.

TL;DR: In this article, the authors present the first textbook that fully explains the neuro-dynamic programming/reinforcement learning methodology, which is a recent breakthrough in the practical application of neural networks and dynamic programming to complex problems of planning, optimal decision making, and intelligent control.

TL;DR: It is shown that options enable temporally abstract knowledge and action to be included in the reinforcement learning frame- work in a natural and general way and may be used interchangeably with primitive actions in planning methods such as dynamic pro- gramming and in learning methodssuch as Q-learning.

TL;DR: PILCO reduces model bias, one of the key problems of model-based reinforcement learning, in a principled way by learning a probabilistic dynamics model and explicitly incorporating model uncertainty into long-term planning.

TL;DR: A new framework to combine tree search with Monte-Carlo evaluation, that does not separate between a min-max phase and a Monte- carlo phase is presented, that provides finegrained control of the tree growth, at the level of individual simulations, and allows efficient selectivity.