Neural Network Dynamics for Model-Based Deep Reinforcement Learning with Model-Free Fine-Tuning

这篇文章的目标是将深度神经网络模型在其他领域中的成功扩展到基于模型的强化学习中。

The contribution of this paper is:

1. They demonstrate effective model-based reinforcement learning with neural network models for several contact-rich simulated locomotion tasks from standard deep reinforcement learning benchmarks.
2. They empirically evaluate a number of design decisions for neural network dynamics model learning.
3. They show how a model-based learner can be used to initialize a model-free learner to achieve high rewards while drastically reducing sample complexity.

Sample Complexity: model-based algorithms>model-free learners

training neural network dynamics models for model-based reinforcement learning
explore how such models can be used to accelerate a model-free learner

model-based acceleration

IV-A detail learned dynamics function
IV-B how to train the learned dynamics function
IV-C how to extract a policy with our learned dynamics function
IV-D how to use reinforcement learning to further improve our learned dynamics function

model-based initialization of model-free reinforcement learning algorithm